gapnium

Tier-2 coefficient · weight 0.20

gapnium.recursion

AI accelerating the creation of better AI.

Now0.1640
12 months+0.0160
Since 2020+0.134

Hover the chart to read values · click a marker to open its articles

What it measures

The degree to which AI systems contribute to AI research and engineering itself: writing the code, kernels and experiments that train their successors, automated ML research, and measurable speed-ups of frontier labs. This is the tier most directly tied to the classic "intelligence explosion" definition of the singularity, so it carries a large weight.

Evidence the sorting agent looks for

  • Share of lab code written by AI
  • Automated research agents (AI Scientist, co-scientist)
  • AI-discovered improvements to training / inference
  • RE-Bench and ML-engineering evals

What 1.0 would mean

AI systems do the large majority of frontier AI research, and each generation measurably shortens the time to the next.

Feedback on this coefficient

Events that moved gapnium.recursion

  1. OpenAI model reports ten more mathematical advances

    OpenAI says an unreleased model made ten new mathematical advances, including solutions to three more Erdős problems.

    Discovery +0.018 Recursion +0.004
    Quanta Magazine · How AI Tore Through a Mathematical Community

  2. Claude Opus 4.5 outscores human engineering candidates

    Anthropic says Opus 4.5 scored higher than any human candidate on its take-home engineering exam, with 80.9% on SWE-bench Verified.

    Autonomy +0.015 Recursion +0.012
    Anthropic · Introducing Claude Opus 4.5

  3. Claude Sonnet 4.5 sustains 30-hour tasks

    Anthropic reports Claude Sonnet 4.5 maintaining focus for more than 30 hours on complex multi-step tasks, and 61% on OSWorld.

    Autonomy +0.020 Recursion +0.006
    Anthropic · Introducing Claude Sonnet 4.5

  4. 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
    Google DeepMind · Gemini achieves gold-medal level at the International Collegiate Programming Contest World Finals

  5. METR RCT: AI tools made experienced developers 19% slower

    A randomised trial finds experienced open-source developers took 19% longer with early-2025 AI tools, despite believing they were 20% faster.

    Automation −0.012 Recursion −0.006
    METR · Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity

  6. AlphaEvolve improves algorithms — and Google's own AI stack

    A Gemini-powered coding agent discovers a faster 4×4 complex matrix multiplication and recovers ~0.7% of Google's worldwide compute.

    Recursion +0.025 Discovery +0.015 Compute +0.005
    Google DeepMind · AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms

  7. Claude 3.7 Sonnet and Claude Code

    Anthropic releases a hybrid reasoning model and Claude Code, an agentic command-line coding tool.

    Autonomy +0.018 Recursion +0.012
    Anthropic · Claude 3.7 Sonnet and Claude Code

  8. Google's AI co-scientist generates validated hypotheses

    A multi-agent Gemini system proposes research hypotheses, including drug-repurposing candidates confirmed in lab experiments.

    Discovery +0.010 Recursion +0.005
    Google Research · Accelerating scientific breakthroughs with an AI co-scientist

  9. RE-Bench: agents vs. human ML researchers

    METR finds AI agents beat human experts on ML research-engineering tasks at 2 hours, but humans win decisively given 8+ hours.

    Recursion +0.010 Autonomy −0.004
    arXiv · RE-Bench: Evaluating frontier AI R&D capabilities of language model agents against human experts

  10. 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
    OpenAI · Learning to reason with LLMs

  11. Sakana's "AI Scientist" writes full ML papers

    An agent pipeline generates ideas, runs ML experiments and writes complete papers for about $15 each.

    Recursion +0.014
    Sakana AI · The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery

  12. 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
    Anthropic · Introducing Claude 3.5 Sonnet

  13. FunSearch makes new discoveries in mathematics

    An LLM-driven evolutionary search finds new constructions for the cap set problem, a first genuine LLM contribution to an open maths problem.

    Discovery +0.012 Recursion +0.006
    Google DeepMind · FunSearch: Making new discoveries in mathematical sciences using Large Language Models

  14. AlphaTensor discovers faster matrix multiplication

    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.

    Discovery +0.012 Recursion +0.010
    Google DeepMind · Discovering novel algorithms with AlphaTensor

  15. Chinchilla rewrites the scaling laws

    DeepMind shows most large models were under-trained; a 70B model on more data beats much larger ones.

    Compute +0.012 Recursion +0.003
    arXiv · Training Compute-Optimal Large Language Models

  16. 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
    Google DeepMind · Competitive programming with AlphaCode

  17. GitHub Copilot puts an AI pair-programmer in the editor

    GitHub previews Copilot, powered by OpenAI Codex, suggesting whole lines and functions as developers type.

    Automation +0.012 Recursion +0.008
    The GitHub Blog · Introducing GitHub Copilot: your AI pair programmer