SciUniverse Level 1
We show this table for reference; we do not rank on it.
SciUniverse Level 1 tests whether models carry out scientific work across chemistry, biology, and materials science. Its 92 tasks include physical experiments, simulated lab work, and measurement analysis, with Pass@1 averaged equally across 17 task families.
Evaluation results published by C5R. Published Level 1 results and evaluation settings. Task contents and model responses are not included.
Pass@1 on SciUniverse Level 1 — October 15, 2026
We mirror the published pass@1 view for SciUniverse Level 1. Claude Fable 5.1 leads the public snapshot at 45.3%, followed by GPT–6 Astra (32.5%) and Claude Opus 5 (30.5%). We do not use these results to rank models overall.
Claude Fable 5.1
Anthropic
C5R facility harness · xhigh reasoning
GPT–6 Astra
OpenAI
C5R facility harness · xhigh reasoning
Claude Opus 5
Anthropic
C5R facility harness · xhigh reasoning
6 model-harness configurationsAgenticCurrentDisplay onlyUpdated October 15, 2026
| Source model | Effort | Overall Pass@1 | Chemistry7 families | Biology5 families | Materials4 families | Cross-domain1 family | Inference cost / attempt | Parameters (B) | Open / closed |
|---|---|---|---|---|---|---|---|---|---|
| Claude Fable 5.1Anthropic | xhigh | 45.3% | 37.4% | 40.0% | 68.8% | 34.4% | $40.61 | Not reported | Closed |
| GPT–6 AstraOpenAI | xhigh | 32.5% | 23.3% | 60.0% | 20.6% | 6.3% | $52.37 | Not reported | Closed |
| Claude Opus 5Anthropic | xhigh | 30.5% | 12.1% | 31.1% | 68.1% | 6.3% | $46.31 | Not reported | Closed |
| Grok 4.6xAI | xhigh | 26.2% | 17.5% | 32.0% | 40.9% | 0.0% | $13.41 | Not reported | Closed |
| Gemini 3.8 FlashGoogle | high | 14.6% | 6.5% | 28.9% | 14.7% | 0.0% | $4.5512 of 361 attempts use estimates | Not reported | Closed |
| GPT–5.6 SolOpenAI | xhigh | 9.4% | 3.1% | 15.6% | 15.3% | 0.0% | $16.53 | Not reported | Closed |
Overall Pass@1 gives each of the 17 task families equal weight. Domain columns average their own families. Inference cost per task attempt includes failures and source estimates, uses the score's family weights, and excludes lab and labor costs. C5R captured the source tariffs on September 9, 2026. Model links are for navigation; these scores stay outside overall and category rankings.
How to read this leaderboard
Overall Pass@1 gives each task family equal weight. Chemistry contains 7 families, biology 5, materials 4, and cross-domain 1; averaging the four domain percentages equally would produce a different score.
Operator receipt: 6 sourced rows are currently displayable on this page; the leading published row is Claude Fable 5.1 at 45.3%.
Honest limit: These results evaluate each model with its agent and facility harness. They stay outside overall and category rankings. Cost covers inference per task attempt, including failures and source estimates, and excludes laboratory and labor costs.
How we show SciUniverse Level 1
The October 15, 2026 snapshot keeps C5R's Level 1 results. The 17 task families receive equal overall weight; chemistry has 7 families, biology 5, materials 4, and cross-domain 1.
The rows evaluate models inside the C5R facility harness. We keep them display only, outside overall and category rankings. Model links are for navigation and do not add these scores to model profiles.
Snapshot
The published SciUniverse Level 1 snapshot places Claude Fable 5.1 first at 45.3%. The third row is 14.8 points behind. The broader top-10 range is 35.9 points, so the table still separates the published systems.
6 model-harness configurations are shown for SciUniverse Level 1. The benchmark falls in the Agentic category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. SciUniverse Level 1 is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About SciUniverse Level 1
Year
2026
Tasks
92 tasks in 17 task families
Format
Scientific work through a facility harness; Pass@1
Difficulty
Level 1 scientific tasks
Models direct equipment and human operators through the C5R facility harness. Level 1 covers work that scientists can complete in a few hours. We show the published overall and four domain scores, effort settings, and inference costs as a standalone table.
Freshness and provenance
Version
SciUniverse Level 1 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does SciUniverse Level 1 measure?
SciUniverse Level 1 tests whether models carry out scientific work across chemistry, biology, and materials science. Its 92 tasks include physical experiments, simulated lab work, and measurement analysis, with Pass@1 averaged equally across 17 task families.
Which model leads the published SciUniverse Level 1 snapshot?
Claude Fable 5.1 currently leads the published SciUniverse Level 1 snapshot with 45.3% pass@1. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on SciUniverse Level 1?
The October 15, 2026 snapshot contains 6 model-harness configurations on 92 tasks in 17 task families.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 5,600+ readers.
One email each week. Unsubscribe anytime.