Benchmark Almanac

OpenAI PRs

GPT-5.2 55% leads since December 2025
Only in system cards

Agents replicate internal OpenAI pull requests from a pre-PR branch and a prompt. Graded by hidden unit tests as pass@1.

In-house OpenAI AI self-improvement evaluation. The tasks are not public.

Made by
OpenAI
Released
January 2025
Kind
Labs' internal R&D tests
Level
Research frontier
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Strongest model tested
GPT-5.2 #50 on the capability index
Last in a system card
December 2025
Last model added
December 2025
12
Models scored in lab cards
13
System cards report it, from 1 lab

2 models share the top score of 55% among models released by December 2025

Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.

The newest model with a score here came out in December 2025, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 55.0 GPT-5.2
  2. 55.0 GPT-5.2 Codex
  3. 53.0 GPT-5.1-Codex-Max
  4. 45.0 GPT-5
  5. 44.0 o3
  6. 42.0 OpenAI deep research
  7. 42.0 GPT-5 mini
  8. 39.0 o4-mini
  9. 37.0 gpt-oss-120b
  10. 37.0 gpt-oss-20b
  11. 7.00 GPT-4.5
  12. 0.0 o3-mini
  1. 1 GPT-5.2 55%
  2. 2 GPT-5.2 Codex 55%
  3. 3 GPT-5.1-Codex-Max 53%
  4. 4 GPT-5 45%
  5. 5 o3 44%
  6. 6 OpenAI deep research 42%
  7. 7 GPT-5 mini 42%
  8. 8 o4-mini 39%
  9. 9 gpt-oss-120b 37%
  10. 10 gpt-oss-20b 37%
  11. 11 GPT-4.5 7.00%
  12. 12 o3-mini 0%

Source: Labs’ model cards

#ModelScoreSettingDateSource
1 GPT-5.2 55% Dec 2025 Lab model card
2 GPT-5.2 Codex 55% Dec 2025 Lab model card
3 GPT-5.1-Codex-Max 53% Nov 2025 Lab model card
4 GPT-5 45% Aug 2025 Lab model card
5 o3 44% Apr 2025 Lab model card
6 OpenAI deep research 42% Feb 2025 Lab model card
7 GPT-5 mini 42% Aug 2025 Lab model card
8 o4-mini 39% Apr 2025 Lab model card
9 gpt-oss-120b 37% Aug 2025 Lab model card
10 gpt-oss-20b 37% Aug 2025 Lab model card
11 GPT-4.5 7.00% Feb 2025 Lab model card
12 o3-mini 0% Jan 2025 Lab model card

6 models have held number one in labs’ own numbers since its release in January 2025

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Lab claims Not tested since Dec 2025
GPT-5.2 55%

Source: Labs’ model cards

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-5.2 GPT-5 o3 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) GPT-5.2 GPT-5

Source: Labs’ model cards

Labs report it only in system cards

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.

13 system cards report it, from 1 lab

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Dec 18, 2025 GPT-5.2 Codex system card addendum 55 OpenAI PRs (no browsing), pass@1, printed chart label (Figure 7, 5.1.3.1) AI R&D risk
Dec 11, 2025 GPT-5.2 system card addendum 55 OpenAI PRs (no browsing), pass@1, printed chart label (Figure 11) AI R&D risk
Nov 18, 2025 GPT-5.1-Codex-Max system card 53 OpenAI PRs (no browsing), pass@1, printed chart label (Figure 13, 5.1.3.4) AI R&D risk
Aug 7, 2025 GPT-5 system card 45 OpenAI PRs (no browsing), pass@1, launch candidate (Figure 20) AI R&D risk
Aug 7, 2025 GPT-5 mini system card 42 OpenAI PRs (no browsing), pass@1, gpt-5-thinking-mini launch candidate (Figure 20) AI R&D risk
Aug 5, 2025 gpt-oss-20b model card 37 OpenAI PRs (no browsing), pass@1, gpt-oss 20b launch candidate (Figure 15) AI R&D risk
Aug 5, 2025 gpt-oss-120b model card 37 OpenAI PRs (no browsing), pass@1, gpt-oss 120b launch candidate (Figure 15) AI R&D risk
Jul 17, 2025 ChatGPT agent system card OpenAI PRs (no browsing), pass@1, chart shows only ChatGPT Agent no-browsing helpful-only at 42%, no released-model value (Figure 17, 5.1.3.3) AI R&D risk
Apr 16, 2025 o4-mini system card 39 OpenAI PRs (no browsing), pass@1, o4-mini launch candidate (Figure 20, 4.4.3) AI R&D risk
Apr 16, 2025 o3 system card 44 OpenAI PRs (no browsing), pass@1, o3 launch candidate (Figure 20, 4.4.3) AI R&D risk
Feb 27, 2025 GPT-4.5 system card 7 OpenAI PRs, pass@1, GPT-4.5 (Post), printed chart label (4.6.5) Autonomy
Feb 25, 2025 OpenAI deep research system card 42 OpenAI PRs, pass@1, post-mitigation without browsing, printed chart label (3.4.7) Autonomy
Jan 31, 2025 o3-mini system card 0 OpenAI PRs, pre- and post-mitigation (5.7.5) Autonomy

More in AI research

All 104 in AI research
  • PaperBench In use Agent replicates ICML 2024 papers from scratch. LLM judge grades against author-written hierarchical rubrics. Qwen 3.8 Max 93%
  • MLS-Bench In use Agent must invent one algorithmic change in an ML research scaffold whose gain transfers across settings. Scored relative to human baselines. Claude Fable 5.1 50.3%
  • PostTrainBench 1.2 In use CLI agent gets a base model, one H100 and a time budget to post-train it for a target benchmark. Scored on normalized gain. Claude Fable 5.1 44.6%
  • MLE-bench In use Kaggle-style ML engineering competitions. Agent trains models and is scored against competition leaderboards. GPT-6 Astra 93.8%
  • KernelBench In use Replace PyTorch operators with faster custom GPU kernels. Scored on correctness and speedup. MiniMax-M3 28.8%
  • PMPP-Hard In use Agents write GPU kernels for 69 tasks that must be correct, and on 48 of them also run within 1.25 times the reference time. The score is the area under the solved-tasks curve over a 10-hour budget, out of 690 task-hours. Inkling 2 Small 494.3