Benchmark Almanac

Research Plan Generation

Qwen-3-4B-RL vs Qwen-3-4B-SFT 97.7% tops its leaderboard
Not in model cards

Models write research plans from goals derived from scientific papers. Judges grade each plan against criteria specific to its source paper.

The December 2025 paper compares GPT, Claude, Gemini, Qwen and other models. GPT-5 Thinking led all eight arXiv subject comparisons, scoring 37.0% to 55.4%.

Made by
Meta
Released
December 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
3
Entries in its paper, 0 models we track

Qwen-3-4B-RL vs Qwen-3-4B-SFT tops the leaderboard with 97.7%

Best score per entry 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. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

  1. 1 Qwen-3-4B-RL vs Qwen-3-4B-SFT ML 97.7%
  2. 2 Qwen-3-4B-RL vs Qwen-3-4B-RL-onlyspec ML 69.1%
  3. 3 Qwen-3-4B-RL vs Qwen-3-4B-RL-onlygen ML 63.4%

Source: The Research Plan Generation paper

#EntryScoreSettingDateSource
1 Qwen-3-4B-RL vs Qwen-3-4B-SFT 97.7% ML Official leaderboard
2 Qwen-3-4B-RL vs Qwen-3-4B-RL-onlyspec 69.1% ML Official leaderboard
3 Qwen-3-4B-RL vs Qwen-3-4B-RL-onlygen 63.4% ML Official 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

All 7 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 pass 4 notes
  • Pass Independent runs Scores from the benchmark’s paper (3 entries).
  • Note Models scored 3 entries on its leaderboard, none a model we track.
  • Note Newest score No scores yet.
  • Note Frontier models No models scored yet.
  • Note Headroom No scores yet.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 3 entries at the last read, none a model we track.
  • 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