Benchmark Almanac

AbGen

DeepSeek-R1 May 2025 4.11 leads since July 2025
Not in model cards

Models design ablation studies to test the contribution of a specified component. The benchmark compares the proposed experiments with expert annotations from 807 NLP papers.

Released
July 2025
Kind
AI research tasks
Level
Expert
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Text
Interaction
Single turn
Items from
Real work and use
Refresh
One fixed set
Run by
Anyone
Strongest model tested
o4-mini #95 on the capability index
Last model added
May 2025
Scores refreshed
November 20th, 2026
18
Models scored in its paper

DeepSeek-R1 leads with 4.11 among models released by May 2025

Best score per model from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best.

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

  1. 1 DeepSeek-R1 May 2025 4.11
  2. 2 GPT-4.1 4.00
  3. 3 o4-mini 4.00
  4. 4 DeepSeek-V3 3.91
  5. 5 Qwen3-235B-A22B 3.90
  6. 6 Gemini 2.5 Flash May 2025 3.86
  7. 7 Gemma 3 27B 3.85
  8. 8 GPT-4o Aug 2024 3.82
  9. 9 Qwen3-32B 3.78
  10. 10 Qwen3-8B 3.74
  11. 11 Mistral Small 3.1 3.64
  12. 12 Phi-4 3.61
  13. 13 Llama 4 Maverick 3.60
  14. 14 deepseek-r1-0528-qwen3-8b 3.51
  15. 15 Qwen2.5-32B 3.47
  16. 16 Llama 4 Scout 3.40
  17. 17 Llama 3.1-70B 3.35
  18. 18 Llama 3.3 70B Instruct 3.28

Source: The AbGen paper

#ModelScoreSettingDateSource
1 DeepSeek-R1 May 2025 4.11 May 2025 Official leaderboard
2 GPT-4.1 4.00 Apr 2025 Official leaderboard
3 o4-mini 4.00 Apr 2025 Official leaderboard
4 DeepSeek-V3 3.91 Dec 2024 Official leaderboard
5 Qwen3-235B-A22B 3.90 Apr 2025 Official leaderboard
6 Gemini 2.5 Flash May 2025 3.86 May 2025 Official leaderboard
7 Gemma 3 27B 3.85 Mar 2025 Official leaderboard
8 GPT-4o Aug 2024 3.82 Aug 2024 Official leaderboard
9 Qwen3-32B 3.78 Apr 2025 Official leaderboard
10 Qwen3-8B 3.74 Apr 2025 Official leaderboard
11 Mistral Small 3.1 3.64 Mar 2025 Official leaderboard
12 Phi-4 3.61 Dec 2024 Official leaderboard
13 Llama 4 Maverick 3.60 Apr 2025 Official leaderboard
14 deepseek-r1-0528-qwen3-8b 3.51 May 2025 Official leaderboard
15 Qwen2.5-32B 3.47 Sep 2024 Official leaderboard
16 Llama 4 Scout 3.40 Apr 2025 Official leaderboard
17 Llama 3.1-70B 3.35 Jul 2024 Official leaderboard
18 Llama 3.3 70B Instruct 3.28 Dec 2024 Official leaderboard

Rankings by part

We read 8 parts from the benchmark’s paper, each ranking models on one part of the test, such as LLM importance, LLM faithfulness and LLM soundness. DeepSeek-R1 leads 5 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

LLM importance

18 models we track have a score here.

  1. 1 Qwen3-235B-A22B ↑4 4.83
  2. 2 GPT-4.1 4.82
  3. 3 Qwen3-32B ↑6 4.82
  4. 4 GPT-4o Aug 2024 ↑4 4.81
  5. 5 o4-mini ↓2 4.80
  6. 6 DeepSeek-R1 May 2025 ↓5 4.80
  7. 7 DeepSeek-V3 ↓3 4.78
  8. 8 Qwen3-8B ↑2 4.77
  9. 9 Phi-4 ↑3 4.74
  10. 10 Mistral Small 3.1 ↑1 4.74

Number one over time

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

DeepSeek
Benchmark paper Not tested since Jul 2025

Source: The AbGen paper

Scores over time

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

Plot with other benchmarks
In the benchmark’s paper
3.25 3.50 3.75 4.00 2025 3.25 3.50 3.75 4.00 2025

Source: Official AbGen leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in AI research

All 164 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-6.2 Astra 96.1%
  • 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. Kimi K3 0.71
  • KernelBench In use Replace PyTorch operators with faster custom GPU kernels. Scored on correctness and speedup. MiniMax-M3 28.8%

Data audit

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 18 months ago, in May 2025, for deepseek-r1-0528-qwen3-8b. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the benchmark’s paper (18 entries).
  • Pass Models scored 18 models we track.
  • Warning Newest score The newest came 18 months ago, in May 2025, for deepseek-r1-0528-qwen3-8b.
  • Note Frontier models None of the top 10 has a score. The strongest tested is o4-mini (#95).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 18 rows for models we track at the last read.
  • Pass Last read November 20th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check