{"generated_at":"2026-11-07T02:51:14+00:00","coefficients":[{"slug":"cognition","symbol":"gapnium.cognition","name":"Cognition","tagline":"Reasoning, knowledge and general problem-solving ability.","weight":0.2,"color":"#3987e5","baseline":0.17,"current":0.572,"change_1y":0.075},{"slug":"autonomy","symbol":"gapnium.autonomy","name":"Autonomy","tagline":"How long and how independently AI agents can work.","weight":0.2,"color":"#d95926","baseline":0.04,"current":0.319,"change_1y":0.097},{"slug":"recursion","symbol":"gapnium.recursion","name":"Recursion","tagline":"AI accelerating the creation of better AI.","weight":0.2,"color":"#199e70","baseline":0.03,"current":0.164,"change_1y":0.016},{"slug":"discovery","symbol":"gapnium.discovery","name":"Discovery","tagline":"The rate at which long-standing problems fall.","weight":0.15,"color":"#c98500","baseline":0.09,"current":0.351,"change_1y":0.073},{"slug":"automation","symbol":"gapnium.automation","name":"Automation","tagline":"Real economic work actually being done by AI.","weight":0.1,"color":"#d55181","baseline":0.06,"current":0.195,"change_1y":0.029},{"slug":"compute","symbol":"gapnium.compute","name":"Compute","tagline":"The physical substrate: chips, data centres, energy.","weight":0.1,"color":"#008300","baseline":0.14,"current":0.263,"change_1y":0.006},{"slug":"embodiment","symbol":"gapnium.embodiment","name":"Embodiment","tagline":"AI acting in the physical world.","weight":0.05,"color":"#9085e9","baseline":0.07,"current":0.116,"change_1y":0}],"points":[{"date":"2020-01-01","overall":0.085,"values":{"cognition":0.17,"autonomy":0.04,"recursion":0.03,"discovery":0.09,"automation":0.06,"compute":0.14,"embodiment":0.07},"events":[]},{"date":"2020-05-28","overall":0.092,"values":{"cognition":0.2,"autonomy":0.04,"recursion":0.03,"discovery":0.09,"automation":0.06,"compute":0.15,"embodiment":0.07},"events":[1]},{"date":"2020-11-30","overall":0.1005,"values":{"cognition":0.205,"autonomy":0.04,"recursion":0.03,"discovery":0.14,"automation":0.06,"compute":0.15,"embodiment":0.07},"events":[2]},{"date":"2021-01-05","overall":0.1021,"values":{"cognition":0.213,"autonomy":0.04,"recursion":0.03,"discovery":0.14,"automation":0.06,"compute":0.15,"embodiment":0.07},"events":[3]},{"date":"2021-06-29","overall":0.1049,"values":{"cognition":0.213,"autonomy":0.04,"recursion":0.038,"discovery":0.14,"automation":0.072,"compute":0.15,"embodiment":0.07},"events":[4]},{"date":"2021-07-15","overall":0.10715,"values":{"cognition":0.213,"autonomy":0.04,"recursion":0.038,"discovery":0.155,"automation":0.072,"compute":0.15,"embodiment":0.07},"events":[5]},{"date":"2022-02-02","overall":0.10995,"values":{"cognition":0.223,"autonomy":0.04,"recursion":0.042,"discovery":0.155,"automation":0.072,"compute":0.15,"embodiment":0.07},"events":[6]},{"date":"2022-03-29","overall":0.11175,"values":{"cognition":0.223,"autonomy":0.04,"recursion":0.045,"discovery":0.155,"automation":0.072,"compute":0.162,"embodiment":0.07},"events":[7]},{"date":"2022-04-04","overall":0.11475,"values":{"cognition":0.235,"autonomy":0.04,"recursion":0.045,"discovery":0.155,"automation":0.072,"compute":0.168,"embodiment":0.07},"events":[8]},{"date":"2022-07-28","overall":0.11775,"values":{"cognition":0.235,"autonomy":0.04,"recursion":0.045,"discovery":0.175,"automation":0.072,"compute":0.168,"embodiment":0.07},"events":[9]},{"date":"2022-10-05","overall":0.12155,"values":{"cognition":0.235,"autonomy":0.04,"recursion":0.055,"discovery":0.187,"automation":0.072,"compute":0.168,"embodiment":0.07},"events":[10]},{"date":"2022-11-30","overall":0.12805,"values":{"cognition":0.245,"autonomy":0.04,"recursion":0.055,"discovery":0.187,"automation":0.107,"compute":0.178,"embodiment":0.07},"events":[11]},{"date":"2023-02-24","overall":0.12905,"values":{"cognition":0.245,"autonomy":0.04,"recursion":0.055,"discovery":0.187,"automation":0.113,"compute":0.182,"embodiment":0.07},"events":[12]},{"date":"2023-03-14","overall":0.14255,"values":{"cognition":0.295,"autonomy":0.05,"recursion":0.055,"discovery":0.187,"automation":0.128,"compute":0.182,"embodiment":0.07},"events":[13]},{"date":"2023-11-29","overall":0.14435,"values":{"cognition":0.295,"autonomy":0.05,"recursion":0.055,"discovery":0.199,"automation":0.128,"compute":0.182,"embodiment":0.07},"events":[14]},{"date":"2023-12-14","overall":0.14735,"values":{"cognition":0.295,"autonomy":0.05,"recursion":0.061,"discovery":0.211,"automation":0.128,"compute":0.182,"embodiment":0.07},"events":[15]},{"date":"2024-01-17","overall":0.15035,"values":{"cognition":0.307,"autonomy":0.05,"recursion":0.061,"discovery":0.215,"automation":0.128,"compute":0.182,"embodiment":0.07},"events":[16]},{"date":"2024-02-15","overall":0.15225,"values":{"cognition":0.315,"autonomy":0.05,"recursion":0.061,"discovery":0.215,"automation":0.128,"compute":0.182,"embodiment":0.076},"events":[17]},{"date":"2024-03-04","overall":0.15525,"values":{"cognition":0.33,"autonomy":0.05,"recursion":0.061,"discovery":0.215,"automation":0.128,"compute":0.182,"embodiment":0.076},"events":[18]},{"date":"2024-03-12","overall":0.15885,"values":{"cognition":0.33,"autonomy":0.068,"recursion":0.061,"discovery":0.215,"automation":0.128,"compute":0.182,"embodiment":0.076},"events":[19]},{"date":"2024-03-18","overall":0.16085,"values":{"cognition":0.33,"autonomy":0.068,"recursion":0.061,"discovery":0.215,"automation":0.128,"compute":0.202,"embodiment":0.076},"events":[20]},{"date":"2024-04-17","overall":0.16135,"values":{"cognition":0.33,"autonomy":0.068,"recursion":0.061,"discovery":0.215,"automation":0.128,"compute":0.202,"embodiment":0.086},"events":[21]},{"date":"2024-05-08","overall":0.16345,"values":{"cognition":0.33,"autonomy":0.068,"recursion":0.061,"discovery":0.229,"automation":0.128,"compute":0.202,"embodiment":0.086},"events":[22]},{"date":"2024-05-13","overall":0.16605,"values":{"cognition":0.338,"autonomy":0.068,"recursion":0.061,"discovery":0.229,"automation":0.138,"compute":0.202,"embodiment":0.086},"events":[23]},{"date":"2024-06-20","overall":0.17325,"values":{"cognition":0.348,"autonomy":0.088,"recursion":0.067,"discovery":0.229,"automation":0.138,"compute":0.202,"embodiment":0.086},"events":[24]},{"date":"2024-07-25","overall":0.17845,"values":{"cognition":0.368,"autonomy":0.088,"recursion":0.067,"discovery":0.237,"automation":0.138,"compute":0.202,"embodiment":0.086},"events":[25]},{"date":"2024-08-13","overall":0.18125,"values":{"cognition":0.368,"autonomy":0.088,"recursion":0.081,"discovery":0.237,"automation":0.138,"compute":0.202,"embodiment":0.086},"events":[26]},{"date":"2024-09-12","overall":0.19205,"values":{"cognition":0.408,"autonomy":0.088,"recursion":0.091,"discovery":0.237,"automation":0.138,"compute":0.21,"embodiment":0.086},"events":[27]},{"date":"2024-10-09","overall":0.19355,"values":{"cognition":0.408,"autonomy":0.088,"recursion":0.091,"discovery":0.247,"automation":0.138,"compute":0.21,"embodiment":0.086},"events":[28]},{"date":"2024-10-22","overall":0.19805,"values":{"cognition":0.408,"autonomy":0.108,"recursion":0.091,"discovery":0.247,"automation":0.143,"compute":0.21,"embodiment":0.086},"events":[29]},{"date":"2024-11-07","overall":0.19605,"values":{"cognition":0.398,"autonomy":0.108,"recursion":0.091,"discovery":0.247,"automation":0.143,"compute":0.21,"embodiment":0.086},"events":[30]},{"date":"2024-11-22","overall":0.19725,"values":{"cognition":0.398,"autonomy":0.104,"recursion":0.101,"discovery":0.247,"automation":0.143,"compute":0.21,"embodiment":0.086},"events":[31]},{"date":"2024-12-20","overall":0.20425,"values":{"cognition":0.433,"autonomy":0.104,"recursion":0.101,"discovery":0.247,"automation":0.143,"compute":0.21,"embodiment":0.086},"events":[32]},{"date":"2025-01-20","overall":0.20755,"values":{"cognition":0.441,"autonomy":0.104,"recursion":0.101,"discovery":0.247,"automation":0.148,"compute":0.222,"embodiment":0.086},"events":[33]},{"date":"2025-01-21","overall":0.21055,"values":{"cognition":0.441,"autonomy":0.104,"recursion":0.101,"discovery":0.247,"automation":0.148,"compute":0.252,"embodiment":0.086},"events":[34]},{"date":"2025-01-23","overall":0.21135,"values":{"cognition":0.433,"autonomy":0.116,"recursion":0.101,"discovery":0.247,"automation":0.148,"compute":0.252,"embodiment":0.086},"events":[35,36]},{"date":"2025-02-02","overall":0.21455,"values":{"cognition":0.433,"autonomy":0.128,"recursion":0.101,"discovery":0.247,"automation":0.156,"compute":0.252,"embodiment":0.086},"events":[37]},{"date":"2025-02-19","overall":0.21705,"values":{"cognition":0.433,"autonomy":0.128,"recursion":0.106,"discovery":0.257,"automation":0.156,"compute":0.252,"embodiment":0.086},"events":[38]},{"date":"2025-02-24","overall":0.22305,"values":{"cognition":0.433,"autonomy":0.146,"recursion":0.118,"discovery":0.257,"automation":0.156,"compute":0.252,"embodiment":0.086},"events":[39]},{"date":"2025-03-19","overall":0.22705,"values":{"cognition":0.433,"autonomy":0.166,"recursion":0.118,"discovery":0.257,"automation":0.156,"compute":0.252,"embodiment":0.086},"events":[40]},{"date":"2025-03-24","overall":0.22505,"values":{"cognition":0.423,"autonomy":0.166,"recursion":0.118,"discovery":0.257,"automation":0.156,"compute":0.252,"embodiment":0.086},"events":[41]},{"date":"2025-03-25","overall":0.22745,"values":{"cognition":0.435,"autonomy":0.166,"recursion":0.118,"discovery":0.257,"automation":0.156,"compute":0.252,"embodiment":0.086},"events":[42]},{"date":"2025-04-16","overall":0.23105,"values":{"cognition":0.445,"autonomy":0.174,"recursion":0.118,"discovery":0.257,"automation":0.156,"compute":0.252,"embodiment":0.086},"events":[43]},{"date":"2025-05-14","overall":0.2388,"values":{"cognition":0.445,"autonomy":0.174,"recursion":0.143,"discovery":0.272,"automation":0.156,"compute":0.257,"embodiment":0.086},"events":[44]},{"date":"2025-05-22","overall":0.2436,"values":{"cognition":0.451,"autonomy":0.192,"recursion":0.143,"discovery":0.272,"automation":0.156,"compute":0.257,"embodiment":0.086},"events":[45]},{"date":"2025-06-07","overall":0.2426,"values":{"cognition":0.446,"autonomy":0.192,"recursion":0.143,"discovery":0.272,"automation":0.156,"compute":0.257,"embodiment":0.086},"events":[46]},{"date":"2025-07-10","overall":0.2402,"values":{"cognition":0.446,"autonomy":0.192,"recursion":0.137,"discovery":0.272,"automation":0.144,"compute":0.257,"embodiment":0.086},"events":[47]},{"date":"2025-07-17","overall":0.2422,"values":{"cognition":0.446,"autonomy":0.202,"recursion":0.137,"discovery":0.272,"automation":0.144,"compute":0.257,"embodiment":0.086},"events":[48]},{"date":"2025-07-21","overall":0.2481,"values":{"cognition":0.471,"autonomy":0.202,"recursion":0.137,"discovery":0.278,"automation":0.144,"compute":0.257,"embodiment":0.086},"events":[49]},{"date":"2025-08-05","overall":0.2494,"values":{"cognition":0.475,"autonomy":0.202,"recursion":0.137,"discovery":0.278,"automation":0.144,"compute":0.257,"embodiment":0.096},"events":[50]},{"date":"2025-08-07","overall":0.2524,"values":{"cognition":0.485,"autonomy":0.202,"recursion":0.137,"discovery":0.278,"automation":0.154,"compute":0.257,"embodiment":0.096},"events":[51]},{"date":"2025-08-18","overall":0.2516,"values":{"cognition":0.485,"autonomy":0.202,"recursion":0.137,"discovery":0.278,"automation":0.146,"compute":0.257,"embodiment":0.096},"events":[52]},{"date":"2025-09-17","overall":0.255,"values":{"cognition":0.497,"autonomy":0.202,"recursion":0.142,"discovery":0.278,"automation":0.146,"compute":0.257,"embodiment":0.096},"events":[53]},{"date":"2025-09-25","overall":0.2576,"values":{"cognition":0.497,"autonomy":0.202,"recursion":0.142,"discovery":0.278,"automation":0.166,"compute":0.257,"embodiment":0.108},"events":[54,55]},{"date":"2025-09-29","overall":0.2628,"values":{"cognition":0.497,"autonomy":0.222,"recursion":0.148,"discovery":0.278,"automation":0.166,"compute":0.257,"embodiment":0.108},"events":[56]},{"date":"2025-10-09","overall":0.2632,"values":{"cognition":0.497,"autonomy":0.222,"recursion":0.148,"discovery":0.278,"automation":0.166,"compute":0.257,"embodiment":0.116},"events":[57]},{"date":"2025-11-13","overall":0.2656,"values":{"cognition":0.497,"autonomy":0.234,"recursion":0.148,"discovery":0.278,"automation":0.166,"compute":0.257,"embodiment":0.116},"events":[58]},{"date":"2025-11-18","overall":0.27,"values":{"cognition":0.519,"autonomy":0.234,"recursion":0.148,"discovery":0.278,"automation":0.166,"compute":0.257,"embodiment":0.116},"events":[59]},{"date":"2025-11-24","overall":0.2754,"values":{"cognition":0.519,"autonomy":0.249,"recursion":0.16,"discovery":0.278,"automation":0.166,"compute":0.257,"embodiment":0.116},"events":[60]},{"date":"2026-01-04","overall":0.27765,"values":{"cognition":0.519,"autonomy":0.249,"recursion":0.16,"discovery":0.293,"automation":0.166,"compute":0.257,"embodiment":0.116},"events":[61]},{"date":"2026-01-22","overall":0.27855,"values":{"cognition":0.519,"autonomy":0.255,"recursion":0.16,"discovery":0.293,"automation":0.163,"compute":0.257,"embodiment":0.116},"events":[62]},{"date":"2026-02-05","overall":0.28475,"values":{"cognition":0.527,"autonomy":0.273,"recursion":0.16,"discovery":0.293,"automation":0.173,"compute":0.257,"embodiment":0.116},"events":[63]},{"date":"2026-04-23","overall":0.28915,"values":{"cognition":0.535,"autonomy":0.287,"recursion":0.16,"discovery":0.293,"automation":0.173,"compute":0.257,"embodiment":0.116},"events":[64]},{"date":"2026-05-20","overall":0.29815,"values":{"cognition":0.55,"autonomy":0.287,"recursion":0.16,"discovery":0.333,"automation":0.173,"compute":0.257,"embodiment":0.116},"events":[65]},{"date":"2026-07-24","overall":0.30695,"values":{"cognition":0.572,"autonomy":0.303,"recursion":0.16,"discovery":0.333,"automation":0.185,"compute":0.257,"embodiment":0.116},"events":[66]},{"date":"2026-08-01","overall":0.31045,"values":{"cognition":0.572,"autonomy":0.303,"recursion":0.164,"discovery":0.351,"automation":0.185,"compute":0.257,"embodiment":0.116},"events":[67]},{"date":"2026-09-22","overall":0.31525,"values":{"cognition":0.572,"autonomy":0.319,"recursion":0.164,"discovery":0.351,"automation":0.195,"compute":0.263,"embodiment":0.116},"events":[68]},{"date":"2026-11-07","overall":0.31525,"values":{"cognition":0.572,"autonomy":0.319,"recursion":0.164,"discovery":0.351,"automation":0.195,"compute":0.263,"embodiment":0.116},"events":[]}],"events":[{"id":1,"slug":"2020-05-gpt-3-shows-few-shot-learning-at-scale","title":"GPT-3 shows few-shot learning at scale","summary":"OpenAI's 175B-parameter GPT-3 performs new tasks from a handful of examples in the prompt, without fine-tuning.","date":"2020-05-28","category":"model_release","significance":4,"impacts":[{"coef":"cognition","delta":0.03,"explanation":"First clear evidence that general task ability emerges from scale alone.","confidence":0.75},{"coef":"compute","delta":0.01,"explanation":"Validated multi-thousand-GPU training runs as a path to capability, triggering investment.","confidence":0.75}],"sources":[{"title":"Language Models are Few-Shot Learners","url":"https://arxiv.org/abs/2005.14165","site":"arXiv","domain":"arxiv.org","date":"2020-05-28"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.007},{"id":2,"slug":"2020-11-alphafold-2-solves-protein-structure-prediction-at-casp14","title":"AlphaFold 2 solves protein structure prediction at CASP14","summary":"DeepMind's AlphaFold 2 predicts protein structures with accuracy competitive with experiment, addressing a 50-year-old grand challenge.","date":"2020-11-30","category":"discovery","significance":5,"impacts":[{"coef":"discovery","delta":0.05,"explanation":"A decades-old grand challenge in biology resolved principally by an AI system.","confidence":0.75},{"coef":"cognition","delta":0.005,"explanation":"Demonstrates deep learning capturing complex physical structure.","confidence":0.75}],"sources":[{"title":"AlphaFold: a solution to a 50-year-old grand challenge in biology","url":"https://deepmind.google/discover/blog/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology/","site":"Google DeepMind","domain":"deepmind.google","date":"2020-11-30"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0085},{"id":3,"slug":"2021-01-dall-e-generates-images-from-text","title":"DALL·E generates images from text","summary":"OpenAI shows a transformer that composes novel images from natural-language descriptions.","date":"2021-01-05","category":"model_release","significance":2,"impacts":[{"coef":"cognition","delta":0.008,"explanation":"Cross-modal compositional generalisation.","confidence":0.75}],"sources":[{"title":"DALL·E: Creating images from text","url":"https://openai.com/index/dall-e/","site":"OpenAI","domain":"openai.com","date":"2021-01-05"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0016},{"id":4,"slug":"2021-06-github-copilot-puts-an-ai-pair-programmer-in-the-editor","title":"GitHub Copilot puts an AI pair-programmer in the editor","summary":"GitHub previews Copilot, powered by OpenAI Codex, suggesting whole lines and functions as developers type.","date":"2021-06-29","category":"deployment","significance":3,"impacts":[{"coef":"automation","delta":0.012,"explanation":"First large-scale deployment of AI into a skilled profession's daily workflow.","confidence":0.75},{"coef":"recursion","delta":0.008,"explanation":"AI starts writing production code, the substrate of AI research itself.","confidence":0.75}],"sources":[{"title":"Introducing GitHub Copilot: your AI pair programmer","url":"https://github.blog/2021-06-29-introducing-github-copilot-ai-pair-programmer/","site":"The GitHub Blog","domain":"github.blog","date":"2021-06-29"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0028},{"id":5,"slug":"2021-07-alphafold-published-in-nature-code-released","title":"AlphaFold published in Nature, code released","summary":"The AlphaFold 2 method is peer-reviewed and open-sourced, and its structure database launches days later.","date":"2021-07-15","category":"research","significance":3,"impacts":[{"coef":"discovery","delta":0.015,"explanation":"AI-generated knowledge becomes a widely used scientific instrument.","confidence":0.75}],"sources":[{"title":"Highly accurate protein structure prediction with AlphaFold","url":"https://www.nature.com/articles/s41586-021-03819-2","site":"Nature","domain":"www.nature.com","date":"2021-07-15"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.00225},{"id":6,"slug":"2022-02-alphacode-reaches-median-competitive-programmer-level","title":"AlphaCode reaches median competitive-programmer level","summary":"DeepMind's AlphaCode ranks around the top 54% of human competitors on Codeforces contests.","date":"2022-02-02","category":"research","significance":2,"impacts":[{"coef":"cognition","delta":0.01,"explanation":"Novel algorithmic problem-solving at human-median level.","confidence":0.75},{"coef":"recursion","delta":0.004,"explanation":"Code-generation ability is a prerequisite for AI-driven AI research.","confidence":0.75}],"sources":[{"title":"Competitive programming with AlphaCode","url":"https://deepmind.google/discover/blog/competitive-programming-with-alphacode/","site":"Google DeepMind","domain":"deepmind.google","date":"2022-02-02"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0028},{"id":7,"slug":"2022-03-chinchilla-rewrites-the-scaling-laws","title":"Chinchilla rewrites the scaling laws","summary":"DeepMind shows most large models were under-trained; a 70B model on more data beats much larger ones.","date":"2022-03-29","category":"research","significance":2,"impacts":[{"coef":"compute","delta":0.012,"explanation":"Effective compute rises without new hardware.","confidence":0.75},{"coef":"recursion","delta":0.003,"explanation":"Better scientific understanding of how to build stronger models.","confidence":0.75}],"sources":[{"title":"Training Compute-Optimal Large Language Models","url":"https://arxiv.org/abs/2203.15556","site":"arXiv","domain":"arxiv.org","date":"2022-03-29"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0018},{"id":8,"slug":"2022-04-palm-scales-to-540b-parameters-with-reasoning-gains","title":"PaLM scales to 540B parameters with reasoning gains","summary":"Google's PaLM shows breakthrough performance on multi-step reasoning with chain-of-thought prompting.","date":"2022-04-04","category":"model_release","significance":2,"impacts":[{"coef":"cognition","delta":0.012,"explanation":"Multi-step reasoning emerges with scale plus chain-of-thought.","confidence":0.75},{"coef":"compute","delta":0.006,"explanation":"Pathways system demonstrates efficient training across thousands of accelerators.","confidence":0.75}],"sources":[{"title":"Pathways Language Model (PaLM): Scaling to 540 Billion Parameters for Breakthrough Performance","url":"https://research.google/blog/pathways-language-model-palm-scaling-to-540-billion-parameters-for-breakthrough-performance/","site":"Google Research","domain":"research.google","date":"2022-04-04"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.003},{"id":9,"slug":"2022-07-alphafold-db-expands-to-200-million-structures","title":"AlphaFold DB expands to 200 million structures","summary":"Predicted structures for nearly every catalogued protein known to science are released freely.","date":"2022-07-28","category":"discovery","significance":3,"impacts":[{"coef":"discovery","delta":0.02,"explanation":"A whole-field dataset produced by AI at a scale beyond human effort.","confidence":0.75}],"sources":[{"title":"AlphaFold reveals the structure of the protein universe","url":"https://deepmind.google/discover/blog/alphafold-reveals-the-structure-of-the-protein-universe/","site":"Google DeepMind","domain":"deepmind.google","date":"2022-07-28"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.003},{"id":10,"slug":"2022-10-alphatensor-discovers-faster-matrix-multiplication","title":"AlphaTensor discovers faster matrix multiplication","summary":"An RL agent finds matrix-multiplication algorithms that beat human-designed ones, including improving a 50-year-old record for 4×4 matrices in a finite field.","date":"2022-10-05","category":"discovery","significance":3,"impacts":[{"coef":"discovery","delta":0.012,"explanation":"Improved a half-century-old algorithmic record.","confidence":0.75},{"coef":"recursion","delta":0.01,"explanation":"AI discovering faster versions of the core operation AI depends on.","confidence":0.75}],"sources":[{"title":"Discovering novel algorithms with AlphaTensor","url":"https://deepmind.google/discover/blog/discovering-novel-algorithms-with-alphatensor/","site":"Google DeepMind","domain":"deepmind.google","date":"2022-10-05"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0038},{"id":11,"slug":"2022-11-chatgpt-launches","title":"ChatGPT launches","summary":"OpenAI releases ChatGPT; it becomes the fastest-growing consumer application in history.","date":"2022-11-30","category":"deployment","significance":5,"impacts":[{"coef":"automation","delta":0.035,"explanation":"Mass adoption: AI becomes a daily tool for knowledge work worldwide.","confidence":0.75},{"coef":"cognition","delta":0.01,"explanation":"RLHF makes model capability usable and steerable.","confidence":0.75},{"coef":"compute","delta":0.01,"explanation":"Triggered the capital surge into AI infrastructure.","confidence":0.75}],"sources":[{"title":"Introducing ChatGPT","url":"https://openai.com/index/chatgpt/","site":"OpenAI","domain":"openai.com","date":"2022-11-30"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0065},{"id":12,"slug":"2023-02-meta-releases-llama-open-weights-spread","title":"Meta releases LLaMA, open weights spread","summary":"Meta's LLaMA models match much larger systems, and their weights quickly spread to the open research community.","date":"2023-02-24","category":"model_release","significance":2,"impacts":[{"coef":"automation","delta":0.006,"explanation":"Capable models become available to anyone to deploy.","confidence":0.75},{"coef":"compute","delta":0.004,"explanation":"Smaller, data-rich models extract more capability per FLOP.","confidence":0.75}],"sources":[{"title":"Introducing LLaMA: A foundational, 65-billion-parameter large language model","url":"https://ai.meta.com/blog/large-language-model-llama-meta-ai/","site":"Meta AI","domain":"ai.meta.com","date":"2023-02-24"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.001},{"id":13,"slug":"2023-03-gpt-4-passes-professional-exams","title":"GPT-4 passes professional exams","summary":"GPT-4 scores around the top 10% on a simulated bar exam and accepts image inputs.","date":"2023-03-14","category":"model_release","significance":5,"impacts":[{"coef":"cognition","delta":0.05,"explanation":"Human-professional-level performance across a broad set of exams.","confidence":0.75},{"coef":"automation","delta":0.015,"explanation":"Capable enough to be integrated into professional workflows.","confidence":0.75},{"coef":"autonomy","delta":0.01,"explanation":"Strong enough to power the first (fragile) autonomous agents.","confidence":0.75}],"sources":[{"title":"GPT-4","url":"https://openai.com/index/gpt-4-research/","site":"OpenAI","domain":"openai.com","date":"2023-03-14"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0135},{"id":14,"slug":"2023-11-gnome-predicts-2-2-million-new-crystals","title":"GNoME predicts 2.2 million new crystals","summary":"DeepMind's GNoME identifies 380,000 stable materials, an order-of-magnitude expansion of known stable inorganic crystals.","date":"2023-11-29","category":"discovery","significance":3,"impacts":[{"coef":"discovery","delta":0.012,"explanation":"Massive expansion of candidate materials, partly validated, partly contested.","confidence":0.75}],"sources":[{"title":"Millions of new materials discovered with deep learning","url":"https://deepmind.google/discover/blog/millions-of-new-materials-discovered-with-deep-learning/","site":"Google DeepMind","domain":"deepmind.google","date":"2023-11-29"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0018},{"id":15,"slug":"2023-12-funsearch-makes-new-discoveries-in-mathematics","title":"FunSearch makes new discoveries in mathematics","summary":"An LLM-driven evolutionary search finds new constructions for the cap set problem, a first genuine LLM contribution to an open maths problem.","date":"2023-12-14","category":"discovery","significance":3,"impacts":[{"coef":"discovery","delta":0.012,"explanation":"First verifiable new mathematical result from an LLM-based system.","confidence":0.75},{"coef":"recursion","delta":0.006,"explanation":"Program-search loop is a template for AI improving algorithms.","confidence":0.75}],"sources":[{"title":"FunSearch: Making new discoveries in mathematical sciences using Large Language Models","url":"https://deepmind.google/discover/blog/funsearch-making-new-discoveries-in-mathematical-sciences-using-large-language-models/","site":"Google DeepMind","domain":"deepmind.google","date":"2023-12-14"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.003},{"id":16,"slug":"2024-01-alphageometry-solves-olympiad-geometry","title":"AlphaGeometry solves olympiad geometry","summary":"AlphaGeometry solves 25 of 30 IMO geometry problems, close to an average gold medallist.","date":"2024-01-17","category":"research","significance":2,"impacts":[{"coef":"cognition","delta":0.012,"explanation":"Olympiad-level mathematical reasoning in a narrow domain.","confidence":0.75},{"coef":"discovery","delta":0.004,"explanation":"Synthetic-data approach to formal mathematics.","confidence":0.75}],"sources":[{"title":"AlphaGeometry: An Olympiad-level AI system for geometry","url":"https://deepmind.google/discover/blog/alphageometry-an-olympiad-level-ai-system-for-geometry/","site":"Google DeepMind","domain":"deepmind.google","date":"2024-01-17"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.003},{"id":17,"slug":"2024-02-sora-generates-minute-long-coherent-video","title":"Sora generates minute-long coherent video","summary":"OpenAI previews Sora, a video model producing up to a minute of high-fidelity video from text.","date":"2024-02-15","category":"model_release","significance":2,"impacts":[{"coef":"cognition","delta":0.008,"explanation":"Large step in generative world modelling.","confidence":0.75},{"coef":"embodiment","delta":0.006,"explanation":"Video models as a possible route to physical understanding.","confidence":0.75}],"sources":[{"title":"Sora: Creating video from text","url":"https://openai.com/index/sora/","site":"OpenAI","domain":"openai.com","date":"2024-02-15"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0019},{"id":18,"slug":"2024-03-claude-3-family-matches-gpt-4","title":"Claude 3 family matches GPT-4","summary":"Anthropic's Claude 3 Opus tops GPT-4 on several reasoning benchmarks including GPQA.","date":"2024-03-04","category":"model_release","significance":3,"impacts":[{"coef":"cognition","delta":0.015,"explanation":"Frontier capability replicated and extended by a second lab.","confidence":0.75}],"sources":[{"title":"Introducing the next generation of Claude","url":"https://www.anthropic.com/news/claude-3-family","site":"Anthropic","domain":"www.anthropic.com","date":"2024-03-04"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.003},{"id":19,"slug":"2024-03-devin-is-billed-as-an-autonomous-ai-software-engineer","title":"Devin is billed as an autonomous AI software engineer","summary":"Cognition unveils Devin, an agent that plans and executes multi-step software tasks, resolving ~14% of SWE-bench issues.","date":"2024-03-12","category":"deployment","significance":2,"impacts":[{"coef":"autonomy","delta":0.018,"explanation":"First widely publicised end-to-end software agent.","confidence":0.75}],"sources":[{"title":"Introducing Devin, the first AI software engineer","url":"https://www.cognition.ai/blog/introducing-devin","site":"Cognition","domain":"www.cognition.ai","date":"2024-03-12"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0036},{"id":20,"slug":"2024-03-nvidia-announces-blackwell","title":"Nvidia announces Blackwell","summary":"Nvidia's Blackwell platform promises large gains in training and inference throughput for trillion-parameter models.","date":"2024-03-18","category":"infrastructure","significance":3,"impacts":[{"coef":"compute","delta":0.02,"explanation":"A new accelerator generation designed explicitly for frontier-scale models.","confidence":0.75}],"sources":[{"title":"NVIDIA Blackwell Platform Arrives to Power a New Era of Computing","url":"https://nvidianews.nvidia.com/news/nvidia-blackwell-platform-arrives-to-power-a-new-era-of-computing","site":"NVIDIA Newsroom","domain":"nvidianews.nvidia.com","date":"2024-03-18"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.002},{"id":21,"slug":"2024-04-boston-dynamics-unveils-all-electric-atlas","title":"Boston Dynamics unveils all-electric Atlas","summary":"The hydraulic Atlas is retired in favour of an all-electric humanoid designed for commercial work.","date":"2024-04-17","category":"robotics","significance":2,"impacts":[{"coef":"embodiment","delta":0.01,"explanation":"Humanoid robotics pivots from research demos toward products.","confidence":0.75}],"sources":[{"title":"An Electric New Era for Atlas","url":"https://bostondynamics.com/blog/electric-new-era-for-atlas/","site":"Boston Dynamics","domain":"bostondynamics.com","date":"2024-04-17"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0005},{"id":22,"slug":"2024-05-alphafold-3-models-all-of-life-s-molecules","title":"AlphaFold 3 models all of life's molecules","summary":"AlphaFold 3 predicts the structure and interactions of proteins, DNA, RNA and ligands.","date":"2024-05-08","category":"discovery","significance":3,"impacts":[{"coef":"discovery","delta":0.014,"explanation":"Extends a solved grand challenge to the interactions relevant for medicine.","confidence":0.75}],"sources":[{"title":"AlphaFold 3 predicts the structure and interactions of all of life's molecules","url":"https://blog.google/technology/ai/google-deepmind-isomorphic-alphafold-3-ai-model/","site":"Google Blog","domain":"blog.google","date":"2024-05-08"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0021},{"id":23,"slug":"2024-05-gpt-4o-brings-real-time-multimodal-interaction","title":"GPT-4o brings real-time multimodal interaction","summary":"OpenAI's GPT-4o reasons across audio, vision and text in real time, at GPT-4 level, free to all users.","date":"2024-05-13","category":"model_release","significance":2,"impacts":[{"coef":"automation","delta":0.01,"explanation":"Frontier capability made free and ubiquitous.","confidence":0.75},{"coef":"cognition","delta":0.008,"explanation":"Native, real-time multimodality.","confidence":0.75}],"sources":[{"title":"Hello GPT-4o","url":"https://openai.com/index/hello-gpt-4o/","site":"OpenAI","domain":"openai.com","date":"2024-05-13"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0026},{"id":24,"slug":"2024-06-claude-3-5-sonnet-raises-the-coding-bar","title":"Claude 3.5 Sonnet raises the coding bar","summary":"Claude 3.5 Sonnet outperforms Claude 3 Opus at a fifth of the cost and solves 64% of problems in an internal agentic coding eval.","date":"2024-06-20","category":"model_release","significance":3,"impacts":[{"coef":"autonomy","delta":0.02,"explanation":"Reliable enough to drive multi-step coding agents.","confidence":0.75},{"coef":"cognition","delta":0.01,"explanation":"Frontier reasoning at much lower cost.","confidence":0.75},{"coef":"recursion","delta":0.006,"explanation":"Widely adopted by engineers, including at AI labs.","confidence":0.75}],"sources":[{"title":"Introducing Claude 3.5 Sonnet","url":"https://www.anthropic.com/news/claude-3-5-sonnet","site":"Anthropic","domain":"www.anthropic.com","date":"2024-06-20"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0072},{"id":25,"slug":"2024-07-alphaproof-and-alphageometry-2-reach-imo-silver","title":"AlphaProof and AlphaGeometry 2 reach IMO silver","summary":"DeepMind's systems solve 4 of 6 problems at the International Mathematical Olympiad, scoring at silver-medal level.","date":"2024-07-25","category":"research","significance":4,"impacts":[{"coef":"cognition","delta":0.02,"explanation":"Elite-level competition mathematics, verified formally.","confidence":0.75},{"coef":"discovery","delta":0.008,"explanation":"Formal theorem proving becomes a practical tool.","confidence":0.75}],"sources":[{"title":"AI achieves silver-medal standard solving International Mathematical Olympiad problems","url":"https://deepmind.google/discover/blog/ai-solves-imo-problems-at-silver-medal-level/","site":"Google DeepMind","domain":"deepmind.google","date":"2024-07-25"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0052},{"id":26,"slug":"2024-08-sakana-s-ai-scientist-writes-full-ml-papers","title":"Sakana's \"AI Scientist\" writes full ML papers","summary":"An agent pipeline generates ideas, runs ML experiments and writes complete papers for about $15 each.","date":"2024-08-13","category":"research","significance":2,"impacts":[{"coef":"recursion","delta":0.014,"explanation":"End-to-end prototype of automated ML research.","confidence":0.75}],"sources":[{"title":"The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery","url":"https://sakana.ai/ai-scientist/","site":"Sakana AI","domain":"sakana.ai","date":"2024-08-13"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0028},{"id":27,"slug":"2024-09-openai-o1-introduces-reasoning-models","title":"OpenAI o1 introduces reasoning models","summary":"o1 \"thinks\" before answering using RL-trained chain of thought, jumping from 13% to 83% on an IMO qualifying exam.","date":"2024-09-12","category":"model_release","significance":5,"impacts":[{"coef":"cognition","delta":0.04,"explanation":"New scaling paradigm; expert-level science and maths reasoning.","confidence":0.75},{"coef":"recursion","delta":0.01,"explanation":"RL on model-generated reasoning: models improving through their own outputs.","confidence":0.75},{"coef":"compute","delta":0.008,"explanation":"Inference-time compute becomes a capability lever.","confidence":0.75}],"sources":[{"title":"Learning to reason with LLMs","url":"https://openai.com/index/learning-to-reason-with-llms/","site":"OpenAI","domain":"openai.com","date":"2024-09-12"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0108},{"id":28,"slug":"2024-10-nobel-prize-in-chemistry-for-alphafold-and-protein-design","title":"Nobel Prize in Chemistry for AlphaFold and protein design","summary":"Demis Hassabis and John Jumper share the Chemistry Nobel with David Baker for computational protein structure prediction and design.","date":"2024-10-09","category":"discovery","significance":3,"impacts":[{"coef":"discovery","delta":0.01,"explanation":"Highest scientific recognition of an AI-produced result.","confidence":0.75}],"sources":[{"title":"The Nobel Prize in Chemistry 2024 — Press release","url":"https://www.nobelprize.org/prizes/chemistry/2024/press-release/","site":"The Nobel Prize","domain":"www.nobelprize.org","date":"2024-10-09"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0015},{"id":29,"slug":"2024-10-claude-gains-computer-use","title":"Claude gains computer use","summary":"Anthropic releases a model that operates a computer by looking at the screen, moving a cursor and typing.","date":"2024-10-22","category":"model_release","significance":3,"impacts":[{"coef":"autonomy","delta":0.02,"explanation":"Agents can act in any GUI, not just through APIs.","confidence":0.75},{"coef":"automation","delta":0.005,"explanation":"Expands the set of tasks that can in principle be automated.","confidence":0.75}],"sources":[{"title":"Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku","url":"https://www.anthropic.com/news/3-5-models-and-computer-use","site":"Anthropic","domain":"www.anthropic.com","date":"2024-10-22"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0045},{"id":30,"slug":"2024-11-frontiermath-frontier-models-solve-under-2","title":"FrontierMath: frontier models solve under 2%","summary":"Epoch AI's research-level maths benchmark shows leading models solving fewer than 2% of problems.","date":"2024-11-07","category":"benchmark","significance":2,"impacts":[{"coef":"cognition","delta":-0.01,"explanation":"Clear evidence of a large gap at research-level mathematics.","confidence":0.75}],"sources":[{"title":"FrontierMath: A Benchmark for Evaluating Advanced Mathematical Reasoning in AI","url":"https://arxiv.org/abs/2411.04872","site":"arXiv","domain":"arxiv.org","date":"2024-11-07"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":-0.002},{"id":31,"slug":"2024-11-re-bench-agents-vs-human-ml-researchers","title":"RE-Bench: agents vs. human ML researchers","summary":"METR finds AI agents beat human experts on ML research-engineering tasks at 2 hours, but humans win decisively given 8+ hours.","date":"2024-11-22","category":"benchmark","significance":2,"impacts":[{"coef":"recursion","delta":0.01,"explanation":"Agents already competitive with experts on short ML-engineering tasks.","confidence":0.75},{"coef":"autonomy","delta":-0.004,"explanation":"Agents scale poorly with longer time budgets.","confidence":0.75}],"sources":[{"title":"RE-Bench: Evaluating frontier AI R&D capabilities of language model agents against human experts","url":"https://arxiv.org/abs/2411.15114","site":"arXiv","domain":"arxiv.org","date":"2024-11-22"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0012},{"id":32,"slug":"2024-12-o3-scores-75-7-on-arc-agi-1","title":"o3 scores 75.7% on ARC-AGI-1","summary":"OpenAI's o3 reaches 75.7% on the ARC-AGI semi-private set (87.5% at high compute), a benchmark that resisted LLMs for five years.","date":"2024-12-20","category":"benchmark","significance":4,"impacts":[{"coef":"cognition","delta":0.035,"explanation":"Large jump on a benchmark explicitly designed to resist memorisation.","confidence":0.75}],"sources":[{"title":"OpenAI o3 Breakthrough High Score on ARC-AGI-Pub","url":"https://arcprize.org/blog/oai-o3-pub-breakthrough","site":"ARC Prize","domain":"arcprize.org","date":"2024-12-20"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.007},{"id":33,"slug":"2025-01-deepseek-r1-open-reasoning-model-at-a-fraction-of-the-cost","title":"DeepSeek-R1: open reasoning model at a fraction of the cost","summary":"DeepSeek releases an open-weights reasoning model performing on par with o1 on maths and code.","date":"2025-01-20","category":"model_release","significance":3,"impacts":[{"coef":"compute","delta":0.012,"explanation":"Big algorithmic-efficiency gain; same capability for far less compute.","confidence":0.75},{"coef":"cognition","delta":0.008,"explanation":"Frontier reasoning reproduced openly.","confidence":0.75},{"coef":"automation","delta":0.005,"explanation":"Cheap, open reasoning models broaden deployment.","confidence":0.75}],"sources":[{"title":"DeepSeek-R1 Release","url":"https://api-docs.deepseek.com/news/news250120","site":"DeepSeek","domain":"api-docs.deepseek.com","date":"2025-01-20"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0033},{"id":34,"slug":"2025-01-stargate-500-billion-ai-infrastructure-project","title":"Stargate: $500 billion AI infrastructure project","summary":"OpenAI, SoftBank, Oracle and MGX announce plans to invest up to $500B in US AI data centres over four years.","date":"2025-01-21","category":"infrastructure","significance":4,"impacts":[{"coef":"compute","delta":0.03,"explanation":"Gigawatt-scale, half-trillion-dollar compute commitment.","confidence":0.75}],"sources":[{"title":"Announcing The Stargate Project","url":"https://openai.com/index/announcing-the-stargate-project/","site":"OpenAI","domain":"openai.com","date":"2025-01-21"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.003},{"id":35,"slug":"2025-01-openai-operator-browses-the-web-for-users","title":"OpenAI Operator browses the web for users","summary":"A browser-using agent that can fill forms, order groceries and book travel is released to users.","date":"2025-01-23","category":"deployment","significance":2,"impacts":[{"coef":"autonomy","delta":0.012,"explanation":"General web agents shipped to consumers.","confidence":0.75}],"sources":[{"title":"Introducing Operator","url":"https://openai.com/index/introducing-operator/","site":"OpenAI","domain":"openai.com","date":"2025-01-23"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0024},{"id":36,"slug":"2025-01-humanity-s-last-exam-models-below-10","title":"Humanity's Last Exam: models below 10%","summary":"A 3,000-question expert-written benchmark finds frontier models scoring under 10%, with poor calibration.","date":"2025-01-23","category":"benchmark","significance":2,"impacts":[{"coef":"cognition","delta":-0.008,"explanation":"New measurement shows a large gap at the frontier of expert knowledge.","confidence":0.75}],"sources":[{"title":"Humanity's Last Exam","url":"https://lastexam.ai/","site":"Humanity's Last Exam","domain":"lastexam.ai","date":"2025-01-23"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":-0.0016},{"id":37,"slug":"2025-02-deep-research-agents-produce-analyst-grade-reports","title":"Deep Research agents produce analyst-grade reports","summary":"OpenAI's Deep Research autonomously browses for up to 30 minutes and writes cited, multi-page reports.","date":"2025-02-02","category":"deployment","significance":2,"impacts":[{"coef":"autonomy","delta":0.012,"explanation":"Long, multi-step research tasks completed unsupervised.","confidence":0.75},{"coef":"automation","delta":0.008,"explanation":"Directly substitutes for analyst research time.","confidence":0.75}],"sources":[{"title":"Introducing deep research","url":"https://openai.com/index/introducing-deep-research/","site":"OpenAI","domain":"openai.com","date":"2025-02-02"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0032},{"id":38,"slug":"2025-02-google-s-ai-co-scientist-generates-validated-hypotheses","title":"Google's AI co-scientist generates validated hypotheses","summary":"A multi-agent Gemini system proposes research hypotheses, including drug-repurposing candidates confirmed in lab experiments.","date":"2025-02-19","category":"research","significance":2,"impacts":[{"coef":"discovery","delta":0.01,"explanation":"AI-generated hypotheses validated experimentally.","confidence":0.75},{"coef":"recursion","delta":0.005,"explanation":"Multi-agent research systems become practical.","confidence":0.75}],"sources":[{"title":"Accelerating scientific breakthroughs with an AI co-scientist","url":"https://research.google/blog/accelerating-scientific-breakthroughs-with-an-ai-co-scientist/","site":"Google Research","domain":"research.google","date":"2025-02-19"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0025},{"id":39,"slug":"2025-02-claude-3-7-sonnet-and-claude-code","title":"Claude 3.7 Sonnet and Claude Code","summary":"Anthropic releases a hybrid reasoning model and Claude Code, an agentic command-line coding tool.","date":"2025-02-24","category":"model_release","significance":3,"impacts":[{"coef":"autonomy","delta":0.018,"explanation":"Agentic coding becomes a mainstream developer workflow.","confidence":0.75},{"coef":"recursion","delta":0.012,"explanation":"Labs report agents writing a large share of their own code.","confidence":0.75}],"sources":[{"title":"Claude 3.7 Sonnet and Claude Code","url":"https://www.anthropic.com/news/claude-3-7-sonnet","site":"Anthropic","domain":"www.anthropic.com","date":"2025-02-24"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.006},{"id":40,"slug":"2025-03-metr-agent-task-length-doubling-every-7-months","title":"METR: agent task length doubling every ~7 months","summary":"METR finds the length of tasks frontier agents complete with 50% reliability has doubled roughly every 7 months for six years.","date":"2025-03-19","category":"study","significance":4,"impacts":[{"coef":"autonomy","delta":0.02,"explanation":"Robust evidence of sustained exponential growth in agent task length.","confidence":0.75}],"sources":[{"title":"Measuring AI Ability to Complete Long Tasks","url":"https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/","site":"METR","domain":"metr.org","date":"2025-03-19"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.004},{"id":41,"slug":"2025-03-arc-agi-2-launches-frontier-models-score-in-single-digits","title":"ARC-AGI-2 launches; frontier models score in single digits","summary":"The harder successor to ARC-AGI sees pure LLMs at 0% and reasoning models at only a few percent, versus ~60% for human panels.","date":"2025-03-24","category":"benchmark","significance":2,"impacts":[{"coef":"cognition","delta":-0.01,"explanation":"Fluid-intelligence gap re-opened by a harder benchmark.","confidence":0.75}],"sources":[{"title":"Announcing ARC-AGI-2 and ARC Prize 2025","url":"https://arcprize.org/blog/announcing-arc-agi-2-and-arc-prize-2025","site":"ARC Prize","domain":"arcprize.org","date":"2025-03-24"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":-0.002},{"id":42,"slug":"2025-03-gemini-2-5-pro-leads-major-leaderboards","title":"Gemini 2.5 Pro leads major leaderboards","summary":"Google's first \"thinking\" Gemini tops LMArena and posts state-of-the-art maths and science scores.","date":"2025-03-25","category":"model_release","significance":2,"impacts":[{"coef":"cognition","delta":0.012,"explanation":"Third lab ships frontier reasoning with very long context.","confidence":0.75}],"sources":[{"title":"Gemini 2.5: Our most intelligent AI model","url":"https://blog.google/technology/google-deepmind/gemini-model-thinking-updates-march-2025/","site":"Google Blog","domain":"blog.google","date":"2025-03-25"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0024},{"id":43,"slug":"2025-04-o3-and-o4-mini-use-tools-while-reasoning","title":"o3 and o4-mini use tools while reasoning","summary":"OpenAI's reasoning models can agentically search, run code and analyse images inside their chain of thought.","date":"2025-04-16","category":"model_release","significance":2,"impacts":[{"coef":"cognition","delta":0.01,"explanation":"Reasoning plus tools in one loop.","confidence":0.75},{"coef":"autonomy","delta":0.008,"explanation":"Models decide for themselves when and how to use tools.","confidence":0.75}],"sources":[{"title":"Introducing OpenAI o3 and o4-mini","url":"https://openai.com/index/introducing-o3-and-o4-mini/","site":"OpenAI","domain":"openai.com","date":"2025-04-16"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0036},{"id":44,"slug":"2025-05-alphaevolve-improves-algorithms-and-google-s-own-ai-stack","title":"AlphaEvolve improves algorithms — and Google's own AI stack","summary":"A Gemini-powered coding agent discovers a faster 4×4 complex matrix multiplication and recovers ~0.7% of Google's worldwide compute.","date":"2025-05-14","category":"discovery","significance":4,"impacts":[{"coef":"recursion","delta":0.025,"explanation":"AI directly sped up training of its own successor models and freed fleet compute.","confidence":0.75},{"coef":"discovery","delta":0.015,"explanation":"Improved a 56-year-old algorithmic record and several open problems.","confidence":0.75},{"coef":"compute","delta":0.005,"explanation":"Efficiency recovered across a hyperscale fleet.","confidence":0.75}],"sources":[{"title":"AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms","url":"https://deepmind.google/discover/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/","site":"Google DeepMind","domain":"deepmind.google","date":"2025-05-14"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.00775},{"id":45,"slug":"2025-05-claude-4-works-autonomously-for-hours","title":"Claude 4 works autonomously for hours","summary":"Claude Opus 4 leads SWE-bench (72.5%) and is reported to run a multi-hour open-source refactor independently.","date":"2025-05-22","category":"model_release","significance":3,"impacts":[{"coef":"autonomy","delta":0.018,"explanation":"Multi-hour unsupervised work in real codebases.","confidence":0.75},{"coef":"cognition","delta":0.006,"explanation":"Incremental frontier gains.","confidence":0.75}],"sources":[{"title":"Introducing Claude 4","url":"https://www.anthropic.com/news/claude-4","site":"Anthropic","domain":"www.anthropic.com","date":"2025-05-22"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0048},{"id":46,"slug":"2025-06-apple-the-illusion-of-thinking","title":"Apple: \"The Illusion of Thinking\"","summary":"Apple researchers report reasoning models collapse to zero accuracy beyond a complexity threshold on controllable puzzles.","date":"2025-06-07","category":"study","significance":2,"impacts":[{"coef":"cognition","delta":-0.005,"explanation":"Evidence (contested) that reasoning gains may not generalise to harder instances.","confidence":0.75}],"sources":[{"title":"The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity","url":"https://machinelearning.apple.com/research/illusion-of-thinking","site":"Apple Machine Learning Research","domain":"machinelearning.apple.com","date":"2025-06-07"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":-0.001},{"id":47,"slug":"2025-07-metr-rct-ai-tools-made-experienced-developers-19-slower","title":"METR RCT: AI tools made experienced developers 19% slower","summary":"A randomised trial finds experienced open-source developers took 19% longer with early-2025 AI tools, despite believing they were 20% faster.","date":"2025-07-10","category":"study","significance":3,"impacts":[{"coef":"automation","delta":-0.012,"explanation":"Rigorous evidence that real productivity lags benchmark capability.","confidence":0.75},{"coef":"recursion","delta":-0.006,"explanation":"Suggests AI speed-up of expert engineering (incl. AI research) was overstated.","confidence":0.75}],"sources":[{"title":"Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity","url":"https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/","site":"METR","domain":"metr.org","date":"2025-07-10"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":-0.0024},{"id":48,"slug":"2025-07-chatgpt-agent-unifies-browsing-research-and-actions","title":"ChatGPT agent unifies browsing, research and actions","summary":"OpenAI merges Operator and Deep Research into one agent with its own virtual computer.","date":"2025-07-17","category":"deployment","significance":2,"impacts":[{"coef":"autonomy","delta":0.01,"explanation":"General-purpose computer agent for mass consumers.","confidence":0.75}],"sources":[{"title":"Introducing ChatGPT agent","url":"https://openai.com/index/introducing-chatgpt-agent/","site":"OpenAI","domain":"openai.com","date":"2025-07-17"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.002},{"id":49,"slug":"2025-07-gemini-deep-think-wins-imo-gold","title":"Gemini Deep Think wins IMO gold","summary":"An advanced Gemini Deep Think model solves 5 of 6 IMO problems in natural language within the time limit, officially certified gold.","date":"2025-07-21","category":"benchmark","significance":4,"impacts":[{"coef":"cognition","delta":0.025,"explanation":"Gold-medal olympiad mathematics by a general language model, under human conditions.","confidence":0.75},{"coef":"discovery","delta":0.006,"explanation":"Brings general models close to research-relevant mathematical reasoning.","confidence":0.75}],"sources":[{"title":"Advanced version of Gemini with Deep Think officially achieves gold-medal standard at the International Mathematical Olympiad","url":"https://deepmind.google/discover/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/","site":"Google DeepMind","domain":"deepmind.google","date":"2025-07-21"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0059},{"id":50,"slug":"2025-08-genie-3-generates-interactive-worlds-in-real-time","title":"Genie 3 generates interactive worlds in real time","summary":"DeepMind's Genie 3 world model generates navigable 3D environments at 24 fps that stay consistent for minutes.","date":"2025-08-05","category":"research","significance":2,"impacts":[{"coef":"embodiment","delta":0.01,"explanation":"World models usable as simulators for embodied agents.","confidence":0.75},{"coef":"cognition","delta":0.004,"explanation":"Learned, consistent physical world modelling.","confidence":0.75}],"sources":[{"title":"Genie 3: A new frontier for world models","url":"https://deepmind.google/discover/blog/genie-3-a-new-frontier-for-world-models/","site":"Google DeepMind","domain":"deepmind.google","date":"2025-08-05"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0013},{"id":51,"slug":"2025-08-gpt-5-released","title":"GPT-5 released","summary":"OpenAI's GPT-5 unifies fast and reasoning modes, posting 74.9% on SWE-bench Verified and lower hallucination rates.","date":"2025-08-07","category":"model_release","significance":3,"impacts":[{"coef":"cognition","delta":0.01,"explanation":"Solid but incremental frontier improvement.","confidence":0.75},{"coef":"automation","delta":0.01,"explanation":"Reasoning-level capability becomes the default for mass users.","confidence":0.75}],"sources":[{"title":"Introducing GPT-5","url":"https://openai.com/index/introducing-gpt-5/","site":"OpenAI","domain":"openai.com","date":"2025-08-07"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.003},{"id":52,"slug":"2025-08-mit-report-95-of-enterprise-gen-ai-pilots-show-no-p-l-impact","title":"MIT report: 95% of enterprise gen-AI pilots show no P&L impact","summary":"An MIT NANDA study finds most corporate generative-AI pilots deliver little measurable financial return.","date":"2025-08-18","category":"study","significance":2,"impacts":[{"coef":"automation","delta":-0.008,"explanation":"Evidence that enterprise adoption lags capability.","confidence":0.75}],"sources":[{"title":"MIT report: 95% of generative AI pilots at companies are failing","url":"https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/","site":"Fortune","domain":"fortune.com","date":"2025-08-18"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":-0.0008},{"id":53,"slug":"2025-09-ai-reaches-gold-at-the-icpc-world-finals","title":"AI reaches gold at the ICPC World Finals","summary":"Gemini 2.5 Deep Think solves 10 of 12 problems at the world's top university programming contest; OpenAI reports 12 of 12.","date":"2025-09-17","category":"benchmark","significance":3,"impacts":[{"coef":"cognition","delta":0.012,"explanation":"Top-of-world algorithmic problem-solving.","confidence":0.75},{"coef":"recursion","delta":0.005,"explanation":"Elite programming capability feeds AI engineering.","confidence":0.75}],"sources":[{"title":"Gemini achieves gold-medal level at the International Collegiate Programming Contest World Finals","url":"https://deepmind.google/discover/blog/gemini-achieves-gold-level-performance-at-the-international-collegiate-programming-contest-world-finals/","site":"Google DeepMind","domain":"deepmind.google","date":"2025-09-17"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0034},{"id":54,"slug":"2025-09-gdpval-models-approach-experts-on-real-occupational-tasks","title":"GDPval: models approach experts on real occupational tasks","summary":"OpenAI's GDPval finds the best models' work rated as good as or better than industry experts' in nearly half of tasks across 44 occupations.","date":"2025-09-25","category":"benchmark","significance":3,"impacts":[{"coef":"automation","delta":0.02,"explanation":"Direct measurement of near-parity on economically valuable tasks.","confidence":0.75}],"sources":[{"title":"Measuring the performance of our models on real-world tasks","url":"https://openai.com/index/gdpval/","site":"OpenAI","domain":"openai.com","date":"2025-09-25"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.002},{"id":55,"slug":"2025-09-gemini-robotics-1-5-brings-agents-into-the-physical-world","title":"Gemini Robotics 1.5 brings agents into the physical world","summary":"DeepMind's vision-language-action models plan multi-step physical tasks and transfer skills across robot bodies.","date":"2025-09-25","category":"robotics","significance":2,"impacts":[{"coef":"embodiment","delta":0.012,"explanation":"Reasoning and cross-embodiment transfer in physical robots.","confidence":0.75}],"sources":[{"title":"Gemini Robotics 1.5 brings AI agents into the physical world","url":"https://deepmind.google/discover/blog/gemini-robotics-15-brings-ai-agents-into-the-physical-world/","site":"Google DeepMind","domain":"deepmind.google","date":"2025-09-25"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0006},{"id":56,"slug":"2025-09-claude-sonnet-4-5-sustains-30-hour-tasks","title":"Claude Sonnet 4.5 sustains 30-hour tasks","summary":"Anthropic reports Claude Sonnet 4.5 maintaining focus for more than 30 hours on complex multi-step tasks, and 61% on OSWorld.","date":"2025-09-29","category":"model_release","significance":3,"impacts":[{"coef":"autonomy","delta":0.02,"explanation":"Day-plus unsupervised runs; large OSWorld jump.","confidence":0.75},{"coef":"recursion","delta":0.006,"explanation":"Agent infrastructure used internally to build AI is opened to everyone.","confidence":0.75}],"sources":[{"title":"Introducing Claude Sonnet 4.5","url":"https://www.anthropic.com/news/claude-sonnet-4-5","site":"Anthropic","domain":"www.anthropic.com","date":"2025-09-29"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0052},{"id":57,"slug":"2025-10-figure-03-designed-for-homes-and-mass-production","title":"Figure 03 designed for homes and mass production","summary":"Figure unveils its third-generation humanoid, built around its Helix VLA model and designed for high-volume manufacturing.","date":"2025-10-09","category":"robotics","significance":2,"impacts":[{"coef":"embodiment","delta":0.008,"explanation":"Humanoids engineered for mass production and domestic use.","confidence":0.75}],"sources":[{"title":"Introducing Figure 03","url":"https://www.figure.ai/news/introducing-figure-03","site":"Figure AI","domain":"www.figure.ai","date":"2025-10-09"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0004},{"id":58,"slug":"2025-11-first-reported-ai-orchestrated-cyber-espionage-campaign","title":"First reported AI-orchestrated cyber-espionage campaign","summary":"Anthropic reports disrupting a state-sponsored campaign in which an AI agent performed 80–90% of the intrusion work autonomously.","date":"2025-11-13","category":"other","significance":3,"impacts":[{"coef":"autonomy","delta":0.012,"explanation":"Real-world, adversarial, largely autonomous multi-step operation.","confidence":0.75}],"sources":[{"title":"Disrupting the first reported AI-orchestrated cyber espionage campaign","url":"https://www.anthropic.com/news/disrupting-AI-espionage","site":"Anthropic","domain":"www.anthropic.com","date":"2025-11-13"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0024},{"id":59,"slug":"2025-11-gemini-3-sets-new-reasoning-records","title":"Gemini 3 sets new reasoning records","summary":"Google's Gemini 3 Pro tops LMArena and posts ~37.5% on Humanity's Last Exam without tools and 31% on ARC-AGI-2.","date":"2025-11-18","category":"model_release","significance":3,"impacts":[{"coef":"cognition","delta":0.022,"explanation":"Large gains on HLE and ARC-AGI-2, benchmarks built to be hard.","confidence":0.75}],"sources":[{"title":"A new era of intelligence with Gemini 3","url":"https://blog.google/products/gemini/gemini-3/","site":"Google Blog","domain":"blog.google","date":"2025-11-18"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0044},{"id":60,"slug":"2025-11-claude-opus-4-5-outscores-human-engineering-candidates","title":"Claude Opus 4.5 outscores human engineering candidates","summary":"Anthropic says Opus 4.5 scored higher than any human candidate on its take-home engineering exam, with 80.9% on SWE-bench Verified.","date":"2025-11-24","category":"model_release","significance":3,"impacts":[{"coef":"autonomy","delta":0.015,"explanation":"New high on real-world software engineering.","confidence":0.75},{"coef":"recursion","delta":0.012,"explanation":"Out-performs human candidates on the lab's own engineering hiring exam.","confidence":0.75}],"sources":[{"title":"Introducing Claude Opus 4.5","url":"https://www.anthropic.com/news/claude-opus-4-5","site":"Anthropic","domain":"www.anthropic.com","date":"2025-11-24"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0054},{"id":61,"slug":"2026-01-erdos-problems-start-falling-to-ai","title":"Erdős problems start falling to AI","summary":"Hobbyists using GPT-5.2 Pro solve Erdős Problem #728, and a DeepMind-led team uses Gemini to resolve four more and locate old solutions to nine.","date":"2026-01-04","category":"discovery","significance":3,"impacts":[{"coef":"discovery","delta":0.015,"explanation":"Multiple genuinely open (if modest) problems resolved, some formally verified.","confidence":0.75}],"sources":[{"title":"How AI Tore Through a Mathematical Community","url":"https://www.quantamagazine.org/how-ai-tore-through-a-mathematical-community-20260803/","site":"Quanta Magazine","domain":"www.quantamagazine.org","date":"2026-08-03"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.00225},{"id":62,"slug":"2026-01-metr-time-horizon-1-1-and-its-limits","title":"METR Time Horizon 1.1 and its limits","summary":"METR expands its task suite to 228 tasks with more than twice as many 8h+ tasks, and cautions that a 50% horizon is not a delegation guarantee.","date":"2026-01-22","category":"benchmark","significance":2,"impacts":[{"coef":"autonomy","delta":0.006,"explanation":"Agents reaching the ceiling of the existing suite.","confidence":0.75},{"coef":"automation","delta":-0.003,"explanation":"Reliability caveats temper delegation claims.","confidence":0.75}],"sources":[{"title":"Time horizon limitations","url":"https://metr.org/notes/2026-01-22-time-horizon-limitations/","site":"METR","domain":"metr.org","date":"2026-01-22"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0009},{"id":63,"slug":"2026-02-claude-opus-4-6-brings-1m-context-to-opus","title":"Claude Opus 4.6 brings 1M context to Opus","summary":"Opus 4.6 posts state-of-the-art results on Terminal-Bench 2.0, Humanity's Last Exam and BrowseComp, and leads GDPval-AA.","date":"2026-02-05","category":"model_release","significance":3,"impacts":[{"coef":"autonomy","delta":0.018,"explanation":"Agent time horizons reach well over a working day.","confidence":0.75},{"coef":"automation","delta":0.01,"explanation":"Leads economically grounded knowledge-work evaluation.","confidence":0.75},{"coef":"cognition","delta":0.008,"explanation":"New highs on HLE.","confidence":0.75}],"sources":[{"title":"Introducing Claude Opus 4.6","url":"https://www.anthropic.com/news/claude-opus-4-6","site":"Anthropic","domain":"www.anthropic.com","date":"2026-02-05"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0062},{"id":64,"slug":"2026-04-gpt-5-5-released","title":"GPT-5.5 released","summary":"OpenAI's GPT-5.5 reports 82.7% on Terminal-Bench 2.0, a large step for terminal-based agentic work.","date":"2026-04-23","category":"model_release","significance":3,"impacts":[{"coef":"autonomy","delta":0.014,"explanation":"Strong gains in agentic terminal work.","confidence":0.75},{"coef":"cognition","delta":0.008,"explanation":"Continued frontier improvement.","confidence":0.75}],"sources":[{"title":"GPT-5.5 — Wikipedia","url":"https://en.wikipedia.org/wiki/GPT-5.5","site":"Wikipedia","domain":"en.wikipedia.org","date":"2026-04-23"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0044},{"id":65,"slug":"2026-05-ai-disproves-erdos-s-1946-unit-distance-conjecture","title":"AI disproves Erdős's 1946 unit distance conjecture","summary":"OpenAI announces an internal reasoning model produced a counterexample to a famous 80-year-old Erdős conjecture, checked by mathematicians including Tim Gowers.","date":"2026-05-20","category":"discovery","significance":5,"impacts":[{"coef":"discovery","delta":0.04,"explanation":"A celebrated, 80-year-old problem resolved with AI as principal contributor.","confidence":0.75},{"coef":"cognition","delta":0.015,"explanation":"Research-level mathematical creativity.","confidence":0.75}],"sources":[{"title":"An AI solution to an 80-year-old problem has shocked mathematicians","url":"https://aihub.org/2026/06/02/an-ai-solution-to-an-80-year-old-problem-has-shocked-mathematicians/","site":"AIhub","domain":"aihub.org","date":"2026-06-02"},{"title":"How AI Tore Through a Mathematical Community","url":"https://www.quantamagazine.org/how-ai-tore-through-a-mathematical-community-20260803/","site":"Quanta Magazine","domain":"www.quantamagazine.org","date":"2026-08-03"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.009},{"id":66,"slug":"2026-07-claude-opus-5-sets-agentic-and-knowledge-work-records","title":"Claude Opus 5 sets agentic and knowledge-work records","summary":"Anthropic's Opus 5 comes close to Claude Fable 5 at half the price and scores roughly three times the next-best model on ARC-AGI 3.","date":"2026-07-24","category":"model_release","significance":4,"impacts":[{"coef":"cognition","delta":0.022,"explanation":"Large lead on an interactive novel-skill benchmark (ARC-AGI 3).","confidence":0.75},{"coef":"autonomy","delta":0.016,"explanation":"State of the art on long-horizon agentic evaluation.","confidence":0.75},{"coef":"automation","delta":0.012,"explanation":"Frontier knowledge-work capability at lower cost per task.","confidence":0.75}],"sources":[{"title":"Introducing Claude Opus 5","url":"https://www.anthropic.com/news/claude-opus-5","site":"Anthropic","domain":"www.anthropic.com","date":"2026-07-24"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0088},{"id":67,"slug":"2026-08-openai-model-reports-ten-more-mathematical-advances","title":"OpenAI model reports ten more mathematical advances","summary":"OpenAI says an unreleased model made ten new mathematical advances, including solutions to three more Erdős problems.","date":"2026-08-01","category":"discovery","significance":3,"impacts":[{"coef":"discovery","delta":0.018,"explanation":"Open problems now falling at a sustained rate rather than one-off.","confidence":0.75},{"coef":"recursion","delta":0.004,"explanation":"Mathematical capability feeds algorithm and architecture design.","confidence":0.75}],"sources":[{"title":"How AI Tore Through a Mathematical Community","url":"https://www.quantamagazine.org/how-ai-tore-through-a-mathematical-community-20260803/","site":"Quanta Magazine","domain":"www.quantamagazine.org","date":"2026-08-03"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0035},{"id":68,"slug":"2026-09-claude-opus-5-5-frontier-performance-at-lower-cost","title":"Claude Opus 5.5: frontier performance at lower cost","summary":"Opus 5.5 reports 66.4% on Terminal-Bench 4.0, 81.8% on OSWorld 2.1 and matches Fable 5.1 on most work for ~40% less than Opus 5.","date":"2026-09-22","category":"model_release","significance":3,"impacts":[{"coef":"autonomy","delta":0.016,"explanation":"Computer-use agents above 80% on OSWorld 2.1.","confidence":0.75},{"coef":"automation","delta":0.01,"explanation":"Frontier capability gets substantially cheaper.","confidence":0.75},{"coef":"compute","delta":0.006,"explanation":"Efficiency: same capability for less inference compute.","confidence":0.75}],"sources":[{"title":"Introducing Claude Opus 5.5","url":"https://www.anthropic.com/news/claude-opus-5-5","site":"Anthropic","domain":"www.anthropic.com","date":"2026-09-22"}],"ai_reviewed_at":"2026-11-07 01:48:50","net":0.0048}],"stats":{"current":0.31525,"gap":0.68475,"change_30d":0,"change_1y":0.05205,"velocity_3y":0.05757,"eta_years":13.2,"eta_year":2039}}