Vals Science Index leaderboard
- GPT-6 Astra leads
- 65.1%
- models measured
- 20
- lead over #2
- +5.5 pts
- median
- 38.3%
Leader on Vals Science Index · read October 22
On October 22, 2026, 20 of our catalog's models had a Vals Science Index result, and OpenAI GPT-6 Astra led with 65.1%, ahead of OpenAI GPT-6.1 Sol at 59.6%.
One measure of agentic science: biology data investigation, mechanism discovery and research workflows, weighted equally by Vals AI.
Role in the index
Vals Science Index is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Cost per test | Duration, as printed | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|---|---|
| 1 | GPT-6 AstraOpenAI | 65.1% | $8.03 | 40m 37s | 85.1 | $50.00 | |
| 2 | GPT-6.1 SolOpenAI | 59.6% | $1.36 | 33m 11s | 82.4 | $10.00 | |
| 3 | Claude Sonnet 5.5Anthropic | 59.1% | $13.24 | 1h 09m | 87.3 | $10.00 | |
| 4 | Claude Opus 5.5Anthropic | 58.7% | $9.13 | 1h 00m | 88.8 | $20.00 | |
| 5 | Gemini 4 ArgonGoogle | 55.4% | $18.40 | 57m 00s | 81.0 | $10.00 | |
| 6 | Claude Opus 5Anthropic | 47.9% | $13.04 | 1h 02m | 83.0 | $25.00 | |
| 7 | GPT-6 SolOpenAI | 45.0% | $2.31 | 31m 16s | 76.0 | $10.00 | |
| 8 | GPT-5.6 SolOpenAI | 41.5% | $3.40 | 43m 32s | 73.5 | $20.00 | |
| 9 | Muse WonderMeta | 39.1% | $8.74 | 43m 08s | 71.9 | $20.00 | |
| 10 | Claude Haiku 5.5Anthropic | 38.5% | $1.86 | 39m 58s | 70.7 | $0.50 | |
| 11 | Grok 4.6xAI | 38.1% | $2.45 | 40m 10s | 65.7 | $6.00 | |
| 12 | Gemini 3.8 FlashGoogle | 35.8% | $2.98 | 22m 00s | 61.6 | $3.75 | |
| 13 | Grok 4.7xAI | 34.8% | $7.12 | 40m 34s | 66.7 | $6.00 | |
| 14 | Ember-1Fireworks | 31.8% | $10.99 | 1h 18m | 52.6 | $15.00 | |
| 15 | MiMo-V2.6-FlashXiaomi | 31.7% | $0.12 | 1h 41m | 54.7 | $0.28 | |
| 16 | DeepSeek V4.1 FlashDeepSeek | 29.6% | $0.29 | 1h 07m | 62.5 | $0.84 | |
| 17 | Step 5 PreviewStepFun | 29.6% | $1.78 | 1h 07m | 61.1 | $2.70 | |
| 18 | GPT-6 LunaOpenAI | 28.4% | $0.11 | 30m 15s | 57.3 | $0.50 | |
| 19 | Gemini 3.1 Pro PreviewGoogle | 27.4% | $2.56 | 18m 03s | 32.3 | $12.00 | |
| 20 | GPT-5.6 LunaOpenAI | 25.3% | $0.26 | 50m 00s | 58.0 | $1.20 |
Only models in our catalog are listed, and each appears once with its best published run. 137 rows were read from the source on October 22, 2026, and rows that match no catalog model are left out. Cost per test is what Vals AI reports one test cost on the run behind the result, in US dollars. Every other column is a figure the board publishes for the same run.