ChemArgus
Models solve chemistry competition and textbook problems with text and figures. A model judge awards partial credit for each subquestion.
- Made by
- Hanyao Guo and colleagues
- Released
- October 2026
- Kind
- Science questions
- Level
- Competition
- Scored
- Graded by a model
- Answers
- Open-ended
- Measures
- Capability
- Subject
- Chemical sciences
- Language
- English
- Interaction
- Single turn
- Items from
- Exams and contests
- Refresh
- One fixed set
- Run by
- A host
- In views
- Vision, Chemistry and materials
- Strongest model tested
- Claude Opus 5 #11 on the capability index
- Last model added
- September 2026
- Scores refreshed
- November 20th, 2026
Claude Opus 5 leads with 64.9% among models released by September 2026
Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.
The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 14 more
- 1 Claude Opus 5 multimodal, oracle_scaffolded 64.9%
- 2
Gemini 3.8 Flash multimodal, oracle_scaffolded 60.3%
- 3 GLM-5.2 text + reader, oracle_scaffolded 52.4%
- 4 Grok 4.6 text, oracle_scaffolded 46%
- 5
Qwen 3.8 Max text, oracle_scaffolded 41.9%
- 6 Kimi K3 text, oracle_scaffolded 37.5%
- 7
DeepSeek-V4-Pro text + reader, oracle_scaffolded 37%
- 8 GPT-5.6 Sol text, oracle_scaffolded 36.5%
- 9
MiniMax-M3 text, oracle_scaffolded 29.1%
Source: Official ChemArgus leaderboard
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5 | 64.9% | multimodal, oracle_scaffolded | Jul 2026 | Official leaderboard |
| 2 | | 60.3% | multimodal, oracle_scaffolded | Sep 2026 | Official leaderboard |
| 3 | GLM-5.2 | 52.4% | text + reader, oracle_scaffolded | Jun 2026 | Official leaderboard |
| 4 | Grok 4.6 | 46% | text, oracle_scaffolded | Aug 2026 | Official leaderboard |
| 5 | | 41.9% | text, oracle_scaffolded | Aug 2026 | Official leaderboard |
| 6 | Kimi K3 | 37.5% | text, oracle_scaffolded | Jul 2026 | Official leaderboard |
| 7 | | 37% | text + reader, oracle_scaffolded | Apr 2026 | Official leaderboard |
| 8 | GPT-5.6 Sol | 36.5% | text, oracle_scaffolded | Jul 2026 | Official leaderboard |
| 9 | | 29.1% | text, oracle_scaffolded | Jun 2026 | Official leaderboard |
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: Official ChemArgus leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Claude Opus 5 | multimodal, oracle_scaffolded | 64.9% | |||
| Claude Opus 5 | text, sequential_carry | 61.7% | |||
| Claude Opus 5 | text, oracle_scaffolded | 61.7% | |||
| Claude Opus 5 | text, full_problem | 61.5% | |||
| multimodal, oracle_scaffolded | 60.3% | ||||
| text, oracle_scaffolded | 59.7% | ||||
| Claude Opus 5 | multimodal, sequential_carry | 59.4% | |||
| text, sequential_carry | 59% | ||||
| multimodal, sequential_carry | 58.5% | ||||
| Claude Opus 5 | text + reader, oracle_scaffolded | 54.7% | |||
| Claude Opus 5 | text + reader, sequential_carry | 53.9% | |||
| Claude Opus 5 | multimodal, full_problem | 53.8% | |||
| GLM-5.2 | text + reader, oracle_scaffolded | 52.4% | |||
| GLM-5.2 | text + reader, sequential_carry | 52.2% | |||
| GLM-5.2 | text + reader, full_problem | 52% | |||
| text + reader, oracle_scaffolded | 49.7% | ||||
| multimodal, full_problem | 48.5% | ||||
| text, full_problem | 46.5% | ||||
| Claude Opus 5 | text + reader, full_problem | 46.2% | |||
| Grok 4.6 | text, oracle_scaffolded | 46% | |||
| Grok 4.6 | multimodal, oracle_scaffolded | 45.5% | |||
| text + reader, sequential_carry | 45.1% | ||||
| Grok 4.6 | text, sequential_carry | 42% | |||
| text, oracle_scaffolded | 41.9% | ||||
| Grok 4.6 | multimodal, sequential_carry | 41.8% | |||
| text, sequential_carry | 39% | ||||
| Kimi K3 | text, oracle_scaffolded | 37.5% | |||
| text + reader, oracle_scaffolded | 37% | ||||
| text + reader, sequential_carry | 36.8% | ||||
| text + reader, full_problem | 36.7% | ||||
| GPT-5.6 Sol | text, oracle_scaffolded | 36.5% | |||
| Kimi K3 | text, sequential_carry | 36% | |||
| text, full_problem | 36% | ||||
| GPT-5.6 Sol | text, sequential_carry | 35.6% | |||
| Kimi K3 | text, full_problem | 34.5% | |||
| GPT-5.6 Sol | text + reader, oracle_scaffolded | 34% | |||
| multimodal, oracle_scaffolded | 34% | ||||
| Grok 4.6 | multimodal, full_problem | 33.8% | |||
| text + reader, full_problem | 33.8% | ||||
| GPT-5.6 Sol | text + reader, sequential_carry | 33.6% | |||
| Grok 4.6 | text, full_problem | 33.4% | |||
| Kimi K3 | text + reader, oracle_scaffolded | 32.4% | |||
| Kimi K3 | text + reader, full_problem | 32.3% | |||
| Kimi K3 | text + reader, sequential_carry | 32.3% | |||
| Kimi K3 | multimodal, oracle_scaffolded | 31.8% | |||
| text + reader, oracle_scaffolded | 31.5% | ||||
| multimodal, sequential_carry | 31.2% | ||||
| GPT-5.6 Sol | text + reader, full_problem | 31.1% | |||
| Grok 4.6 | text + reader, oracle_scaffolded | 31.1% | |||
| Kimi K3 | multimodal, sequential_carry | 30.6% | |||
| GPT-5.6 Sol | multimodal, oracle_scaffolded | 30.4% | |||
| GPT-5.6 Sol | text, full_problem | 29.8% | |||
| multimodal, full_problem | 29.3% | ||||
| text + reader, sequential_carry | 29.2% | ||||
| text, oracle_scaffolded | 29.1% | ||||
| Grok 4.6 | text + reader, sequential_carry | 29% | |||
| GPT-5.6 Sol | multimodal, sequential_carry | 28.8% | |||
| text, sequential_carry | 28.7% | ||||
| Grok 4.6 | text + reader, full_problem | 28.6% | |||
| multimodal, oracle_scaffolded | 27.9% | ||||
| text + reader, full_problem | 27.9% | ||||
| text, oracle_scaffolded | 27.6% | ||||
| multimodal, sequential_carry | 27.5% | ||||
| Kimi K3 | multimodal, full_problem | 27.5% | |||
| multimodal, full_problem | 27.3% | ||||
| text + reader, sequential_carry | 27.3% | ||||
| text + reader, oracle_scaffolded | 27.3% | ||||
| text + reader, full_problem | 27.2% | ||||
| text, full_problem | 26.2% | ||||
| GLM-5.2 | text, oracle_scaffolded | 25.1% | |||
| text, full_problem | 25% | ||||
| text, sequential_carry | 25% | ||||
| GLM-5.2 | text, sequential_carry | 24.9% | |||
| GLM-5.2 | text, full_problem | 23.5% | |||
| GPT-5.6 Sol | multimodal, full_problem | 23.2% |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official ChemArgus leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official ChemArgus 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 Science and reasoning
- GPQA Diamond In use 198 graduate-level 'Google-proof' multiple-choice questions in biology, chemistry and physics.
Gemini 4 Carbon 97%
- Terminal-Bench-Science In use 70 scientific research-workflow tasks across five science areas, run in Harbor containers. Hidden tests give the resolution rate. GPT-6 Astra 68.1%
- BioMysteryBench In use Bioinformatics problems on anonymized real datasets that require real analysis with tools. Rubric-graded, with human-solvable and human-difficult splits. Claude Fable 5.2 83.7%
- FrontierScience In use Models solve physics, chemistry, and biology problems in an Olympiad track graded by short answer and a Research track graded by 10-point rubrics.
Intern-S2-397B 87%
- HiPhO In use Models solve theory problems from 2024 and 2025 physics olympiad exams, some with diagrams. Official marking schemes grade answers and steps, mapped to medal levels.
Gemini 3 Pro 78.3%
- LABBench2 In use 11 biology research sub-tasks (literature, figures, tables, patents, trials and more) with tool access. Macro-averaged accuracy. GPT-5.2 60.9%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Task access check. We don’t know whether the tasks are public. Every benchmark
Catalog record 1 warning 1 pass
- Warning Task access We don’t know whether the tasks are public.
- Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
- Pass Independent runs Scores from the official leaderboard (9 entries).
- Pass Models scored 9 models we track.
- Pass Newest score The newest is from September 2026, for Gemini 3.8 Flash.
- Note Frontier models None of the top 10 has a score. The strongest tested is Claude Opus 5 (#11).
- Pass Headroom The best is 64.9%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on cosmosmind.ai: 75 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.