Internal model use survey (Anthropic)
Anthropic technical staff estimate productivity uplift from the model and judge whether it could automate a junior ML researcher.
Survey of Anthropic staff, not a task set.
- Made by
- Anthropic
- Released
- May 2025
- Kind
- Labs' internal R&D tests
- Level
- Expert
- Scored
- Graded by people
- Answers
- Open-ended
- Tasks
- Kept private
- Strongest model tested
- Claude Opus 4.6 #37 on the capability index
- Last in a system card
- April 2026
- Last model added
- April 2026
Claude Opus 4.5 leads with 50% among models released by April 2026
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 April 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- Claude Fable 5.2
-
Gemini 4 Carbon - GPT-6 Astra
- Claude Opus 5.5
- and 16 more
- 1 Claude Opus 4.5 50%
- 2 Claude Mythos Preview 5.60%
- 3 Claude Opus 4 0%
- 4 Claude Sonnet 4.5 0%
- 5 Claude Opus 4.6 0%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.5 | 50% | Nov 2025 | Lab model card | |
| 2 | Claude Mythos Preview | 5.60% | Apr 2026 | Lab model card | |
| 3 | Claude Opus 4 | 0% | May 2025 | Lab model card | |
| 4 | Claude Sonnet 4.5 | 0% | Sep 2025 | Lab model card | |
| 5 | Claude Opus 4.6 | 0% | Feb 2026 | Lab model card |
2 models have held number one in labs’ own numbers since its release in May 2025
Lab claims count from the day the card came out, independent runs from the model’s release.
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.
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.
5 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.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Apr 7, 2026 | Claude Mythos Preview system card | 5.6 | Internal survey (n=18), 1/18 thought the model already was a drop-in replacement for an entry-level Research Scientist or Engineer, 4/18 gave it 50% odds with 3 months of scaffolding (2.3.4). A separate staff survey gives a productivity uplift geometric mean of about 4x (2.3.6) | AI R&D risk |
| Feb 5, 2026 | Claude Opus 4.6 system card | 0 | Internal model use survey, 0/16 participants believed the model could be a drop-in replacement for an entry-level researcher after follow-up (initial answers 11 unlikely, 3 likely, 2 already possible); productivity uplift mean 152%, median 100% (8.3, 8.3.1) | AI R&D risk |
| Nov 24, 2025 | Claude Opus 4.5 system card | 50 | Internal model use survey, 9/18 participants reported at least 100% productivity improvement, median 100%, mean 220%, 0/18 believed AI R&D-4 was crossed (7.3.4) | AI R&D risk |
| Sep 29, 2025 | Claude Sonnet 4.5 system card | 0 | Internal model evaluation and use survey, 0/7 researchers believed the model could completely automate a junior ML researcher. Productivity estimates ~100%, 15%, 20%, 20%, 30%, 40% (9.3.5) | AI R&D risk |
| May 22, 2025 | Claude Opus 4 system card | 0 | Internal model use survey, 0/4 researchers believed Claude Opus 4 could completely automate a junior ML researcher (7.3.5) | AI R&D risk |
More 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