Tier-2 coefficient · weight 0.20
gapnium.cognition
Reasoning, knowledge and general problem-solving ability.
Hover the chart to read values · click a marker to open its articles
What it measures
How close frontier models are to matching or exceeding the best humans across reasoning, knowledge, mathematics, language and multimodal understanding. Driven mainly by results on hard, uncontaminated benchmarks (GPQA, FrontierMath, Humanity's Last Exam, ARC-AGI), competition performance (IMO, ICPC) and qualitative capability jumps.
Evidence the sorting agent looks for
- Frontier benchmark scores (GPQA, HLE, FrontierMath, ARC-AGI)
- Olympiad / competition results
- New modalities and generalisation
- Evidence of benchmark saturation or failure
What 1.0 would mean
Systems reliably exceed the best human experts in every cognitive domain, including novel problems no benchmark anticipated.
Events that moved gapnium.cognition
-
Claude Opus 5 sets agentic and knowledge-work records
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.
Cognition +0.022 Autonomy +0.016 Automation +0.012 -
AI disproves Erdős's 1946 unit distance conjecture
OpenAI announces an internal reasoning model produced a counterexample to a famous 80-year-old Erdős conjecture, checked by mathematicians including Tim Gowers.
Discovery +0.040 Cognition +0.015 -
GPT-5.5 released
OpenAI's GPT-5.5 reports 82.7% on Terminal-Bench 2.0, a large step for terminal-based agentic work.
Autonomy +0.014 Cognition +0.008 -
Claude Opus 4.6 brings 1M context to Opus
Opus 4.6 posts state-of-the-art results on Terminal-Bench 2.0, Humanity's Last Exam and BrowseComp, and leads GDPval-AA.
Autonomy +0.018 Automation +0.010 Cognition +0.008 -
Gemini 3 sets new reasoning records
Google's Gemini 3 Pro tops LMArena and posts ~37.5% on Humanity's Last Exam without tools and 31% on ARC-AGI-2.
Cognition +0.022 -
AI reaches gold at the ICPC World Finals
Gemini 2.5 Deep Think solves 10 of 12 problems at the world's top university programming contest; OpenAI reports 12 of 12.
Cognition +0.012 Recursion +0.005 -
GPT-5 released
OpenAI's GPT-5 unifies fast and reasoning modes, posting 74.9% on SWE-bench Verified and lower hallucination rates.
Cognition +0.010 Automation +0.010 -
Genie 3 generates interactive worlds in real time
DeepMind's Genie 3 world model generates navigable 3D environments at 24 fps that stay consistent for minutes.
Embodiment +0.010 Cognition +0.004 -
Gemini Deep Think wins IMO gold
An advanced Gemini Deep Think model solves 5 of 6 IMO problems in natural language within the time limit, officially certified gold.
Cognition +0.025 Discovery +0.006 -
Apple: "The Illusion of Thinking"
Apple researchers report reasoning models collapse to zero accuracy beyond a complexity threshold on controllable puzzles.
Cognition −0.005 -
Claude 4 works autonomously for hours
Claude Opus 4 leads SWE-bench (72.5%) and is reported to run a multi-hour open-source refactor independently.
Autonomy +0.018 Cognition +0.006 -
o3 and o4-mini use tools while reasoning
OpenAI's reasoning models can agentically search, run code and analyse images inside their chain of thought.
Cognition +0.010 Autonomy +0.008 -
Gemini 2.5 Pro leads major leaderboards
Google's first "thinking" Gemini tops LMArena and posts state-of-the-art maths and science scores.
Cognition +0.012 -
ARC-AGI-2 launches; frontier models score in single digits
The harder successor to ARC-AGI sees pure LLMs at 0% and reasoning models at only a few percent, versus ~60% for human panels.
Cognition −0.010 -
Humanity's Last Exam: models below 10%
A 3,000-question expert-written benchmark finds frontier models scoring under 10%, with poor calibration.
Cognition −0.008 -
DeepSeek-R1: open reasoning model at a fraction of the cost
DeepSeek releases an open-weights reasoning model performing on par with o1 on maths and code.
Compute +0.012 Cognition +0.008 Automation +0.005 -
o3 scores 75.7% on ARC-AGI-1
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.
Cognition +0.035 -
FrontierMath: frontier models solve under 2%
Epoch AI's research-level maths benchmark shows leading models solving fewer than 2% of problems.
Cognition −0.010 -
OpenAI o1 introduces reasoning models
o1 "thinks" before answering using RL-trained chain of thought, jumping from 13% to 83% on an IMO qualifying exam.
Cognition +0.040 Recursion +0.010 Compute +0.008 -
AlphaProof and AlphaGeometry 2 reach IMO silver
DeepMind's systems solve 4 of 6 problems at the International Mathematical Olympiad, scoring at silver-medal level.
Cognition +0.020 Discovery +0.008 -
Claude 3.5 Sonnet raises the coding bar
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.
Autonomy +0.020 Cognition +0.010 Recursion +0.006 -
GPT-4o brings real-time multimodal interaction
OpenAI's GPT-4o reasons across audio, vision and text in real time, at GPT-4 level, free to all users.
Automation +0.010 Cognition +0.008 -
Claude 3 family matches GPT-4
Anthropic's Claude 3 Opus tops GPT-4 on several reasoning benchmarks including GPQA.
Cognition +0.015 -
Sora generates minute-long coherent video
OpenAI previews Sora, a video model producing up to a minute of high-fidelity video from text.
Cognition +0.008 Embodiment +0.006 -
AlphaGeometry solves olympiad geometry
AlphaGeometry solves 25 of 30 IMO geometry problems, close to an average gold medallist.
Cognition +0.012 Discovery +0.004 -
GPT-4 passes professional exams
GPT-4 scores around the top 10% on a simulated bar exam and accepts image inputs.
Cognition +0.050 Automation +0.015 Autonomy +0.010 -
ChatGPT launches
OpenAI releases ChatGPT; it becomes the fastest-growing consumer application in history.
Automation +0.035 Cognition +0.010 Compute +0.010 -
PaLM scales to 540B parameters with reasoning gains
Google's PaLM shows breakthrough performance on multi-step reasoning with chain-of-thought prompting.
Cognition +0.012 Compute +0.006 -
AlphaCode reaches median competitive-programmer level
DeepMind's AlphaCode ranks around the top 54% of human competitors on Codeforces contests.
Cognition +0.010 Recursion +0.004 -
DALL·E generates images from text
OpenAI shows a transformer that composes novel images from natural-language descriptions.
Cognition +0.008 -
AlphaFold 2 solves protein structure prediction at CASP14
DeepMind's AlphaFold 2 predicts protein structures with accuracy competitive with experiment, addressing a 50-year-old grand challenge.
Discovery +0.050 Cognition +0.005 -
GPT-3 shows few-shot learning at scale
OpenAI's 175B-parameter GPT-3 performs new tasks from a handful of examples in the prompt, without fine-tuning.
Cognition +0.030 Compute +0.010