# STEMBench leaderboard

Reasoning benchmark

> Every figure below is reproduced as its upstream source published it: nothing is modelled, estimated, interpolated or converted. `Unavailable` means no source published the value — it is never a zero. Each value carries its evidence state and the date it was observed.

Page: https://modelscale.dev/leaderboards/benchmarks/reasoning-stem-bench120

STEMBench120: six-domain scientific research progress. Every model the catalog carries a published STEMBench value for, ranked by that value.

## Benchmark record

| Field | Value |
| --- | --- |
| Category | Reasoning |
| Measure | Audited progress level 0 to 4 per task, refined to a fractional score; complete-run mean across all tasks |
| Tasks | 120 research tasks, 20 in each of six STEM domains |
| Difficulty | Open research targets; no run has reached a full-target success |
| Published by | [STEMBench: six-domain model evaluation](https://www.ulam.ai/stembench) |

Scores how far a model advances 120 open research tasks across biology, chemistry, physics, astronomy, Earth science and engineering on a 0 to 4 progress scale, crediting supported partial work and honest statements of what remains unresolved.

## STEMBench ranking

9 models with a published STEMBench value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.

| Rank | Model | Provider | Audited progress level 0 to 4 per task, refined to a fractional score; complete-run mean across all tasks | Evidence |
| --- | --- | --- | --- | --- |
| 1 | [GPT-6 Astra](https://modelscale.dev/models/gpt-6-astra) | OpenAI | 2.81 | Observed 2026-10-23 · source benchlm:models |
| 2 | [GPT-6.1 Sol](https://modelscale.dev/models/gpt-6-1-sol) | OpenAI | 2.76 | Observed 2026-10-23 · source benchlm:models |
| 3 | [Muse Wonder](https://modelscale.dev/models/muse-wonder) | Meta | 2.68 | Observed 2026-10-23 · source benchlm:models |
| 4 | [Claude Opus 5.5](https://modelscale.dev/models/claude-opus-5-5) | Anthropic | 2.62 | Observed 2026-10-23 · source benchlm:models |
| 5 | [Hy4 preview](https://modelscale.dev/models/hy4-preview) | Tencent | 2.59 | Observed 2026-10-23 · source benchlm:models |
| 6 | [Mistral Large 4](https://modelscale.dev/models/mistral-large-4) | Mistral | 2.50 | Observed 2026-10-23 · source benchlm:models |
| 7 | [DeepSeek V4.1 Flash 1017](https://modelscale.dev/models/deepseek-v4-1-flash) | DeepSeek | 2.49 | Observed 2026-10-23 · source benchlm:models |
| 8 | [Gemini 4 Argon](https://modelscale.dev/models/gemini-4-argon) | Google | 2.44 | Observed 2026-10-23 · source benchlm:models |
| 9 | [MiMo-V2.6-Pro](https://modelscale.dev/models/mimo-v2-6-pro) | Xiaomi | 2.17 | Observed 2026-10-23 · source benchlm:models |

Rows are ordered by the value the source published, highest first. The source does not state whether a higher value is the better result, so this document does not either: for a benchmark that measures a rate of failure, read the table from the bottom.