Proprietary

Vals Science Index

Updated 11/19/2026

A single measure of AI's ability to do science — agentic model performance on biology data investigation, mechanism discovery, and research workflows, weighted equally.

As of November 19, 2026, GPT-6.2 Astra ranks first on the Vals Science Index with 68.44%, followed by GPT-6 Astra (65.09%) and Claude Fable 5.2 (63.74%).

Vals Science IndexAgentic science benchmark
Vals Science Index•Nov 19, 2026
ACCURACY
Lab:
Pinch or Ctrl/Cmd + scroll to zoom. Once zoomed, scroll or swipe with two fingers to pan, or drag to pan. Shift + drag draws a box to zoom into; turn on box zoom to draw one without Shift. Plus and minus keys zoom, arrow keys pan, 0 resets.

Vals Science Index leaderboard

Rank Model Accuracy Cost / Test Tokens / Test Duration / Test
1 GPT-6.2 Astra 68.44% $4.76 3.48m 22m15s
2 GPT-6 Astra 65.09% $8.03 2.19m 40m37s
3 Claude Fable 5.2 63.74% $8.79 15.32m 34m41s
4 Gemini 4 Neon 62.63% $8.26 16.16m 40m52s
5 GPT-6.1 Sol 59.63% $1.36 1.91m 33m11s
6 Claude Sonnet 5.5 59.12% $8.50 48.02m 1h09m
7 Claude Opus 5.5 58.65% $8.71 11.18m 59m51s
8 GPT-6.2 Sol 57.76% $1.12 3.21m 19m00s
9 Gemini 4 Argon 55.36% $18.40 12.87m 57m00s
10 Wonder Full 48.54% $10.15 21.52m 1h34m
11 Claude Opus 5 47.93% $13.04 - 1h02m
12 GPT-6 Sol 45.00% $2.31 5.92m 31m16s
13 Kimi K3.1 Pro 44.37% $3.33 11.83m 23m02s
14 GPT-5.6 Sol 41.48% $3.40 - 43m32s
15 Wonder 39.12% $8.74 24.90m 43m08s
16 Claude Haiku 5.5 38.50% $1.86 26.15m 39m58s
17 Grok 4.6 38.09% $2.45 3.19m 40m10s
18 Gemini 3.8 Flash 35.76% $2.98 9.25m 22m00s
19 Grok 4.7 34.83% $7.12 7.07m 40m34s
20 Inkling 2 Small 32.69% $1.99 70.74m 57m29s
21 Ember-1 31.76% $10.99 27.35m 1h18m
22 MiMo V2.6 Flash 31.72% $0.12 12.14m 1h41m
23 DeepSeek V4.1 Flash 29.65% $0.29 - 1h07m
24 Step 5 Preview 29.58% $1.78 21.39m 1h07m
25 GPT-6 Luna 28.38% $0.11 4.43m 30m15s
26 Gemini 3.1 Pro Preview (02/26) 27.38% $2.56 4.80m 18m03s
27 GPT-5.6 Luna 25.30% $0.26 - 50m00s
28 Inkling 2 Mini 23.58% $1.66 51.51m 45m29s
29 Beam 1.0 22.03% - 11.09m 29m08s

Motivation

AI systems have tremendous potential to advance science. Unlike data from enterprise contexts, scientific data can be collected without coordinating large groups of people. Because ground truth verification in science domains is less dependent on human authorship, science evaluations also allow us to characterize superhuman capabilities. In BioMysteryBench, for example, a subset of tasks is explicitly not solvable by human experts. Many science tasks require the same underlying skills as coding and math tasks, but additionally require niche domain knowledge and discernment. The ultimate promise of expanding the science capabilities of AI systems is in the material improvement of the human condition via fundamental breakthroughs and discovery. Here, we make an effort to monitor advances in these capabilities.

Most science benchmarks test recall. The three benchmarks in this index give a model a dataset, a simulated lab, or a terminal and score what it finds:

  • BioMysteryBench: investigating real biological datasets to answer open questions
  • MysteryMechanism: recovering hidden mechanisms by running experiments
  • Terminal-Bench-Science: research workflows across the life, physical, earth, mathematical, and engineering sciences

The Vals Science Index averages the three into one number, so models can be compared on research work as a whole rather than on one field or task format.


Leaders by Discipline

No one model leads every science

The index's top five, ranked among all 29 index models.

Component GPT-6.2 Astra GPT-6 Astra Claude Fable 5.2 Gemini 4 Neon GPT-6.1 Sol
Index #1 #2 #3 #4 #5
Benchmarks
BioMystery #3 #6 #1 #4 #5
Mechanism #2 #1 #3 #8 #6
TB Science #1 #3 #5 #2 #6
TB Science disciplines
Life #2 #3 #1 #4 #7
Physical #3 #8 #4 #2 #10
Earth #2 #7 #9 #1 #5
Math #1 #2 #9 #5 #3
Engineering #1 #7 #6 #5 #4
  • #1
  • Top 5
  • Top half
  • Bottom half

Vals Science Index vs Vals Index

The science order is not the Vals Index order

Leaderboard rank on the Vals Index (left) and the Vals Science Index (right), for the models in the top 10 of both.

Vals Index Vals Science Index #8 GPT-6.2 Astra GPT-6.2 Astra #1 #1 Claude Fable 5.2 Claude Fable 5.2 #3 #2 Gemini 4 Neon Gemini 4 Neon #4 #3 Claude Sonnet 5.5 Claude Sonnet 5.5 #6 #5 Claude Opus 5.5 Claude Opus 5.5 #7 #10 GPT-6.2 Sol GPT-6.2 Sol #8 #6 Gemini 4 Argon Gemini 4 Argon #9 #9 Wonder Full Wonder Full #10

Methodology

Formula

The Vals Science Index is an even weighting of three benchmarks: BioMysteryBench, MysteryMechanism, and Terminal-Bench-Science.

Science_Index = AVG(BioMysteryBench, MysteryMechanism, TerminalBenchScience)

Updates

10/16/2026

Released Vals Science Index v1: the average of BioMysteryBench, MysteryMechanism, and Terminal-Bench-Science.