Vals AI (Vals Index): leaderboard
A composite index rather than one task: Vals AI averages its finance benchmarks (Finance Agent v2 and Excel Modeling), its coding benchmarks (Terminal-Bench 2.1, Vibe Code Bench and Code Migration) and its legal benchmarks (Legal Research Bench and HLAB), then weights the three sector averages by each sector's share of US GDP. The parts are agentic runs on mostly private sets, higher is better.
Metric: Accuracy (%). Source: www.vals.ai. Saturation forecast: Around June 2027. 50 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 4 Argon | 68.9 | #8 |
| 2 | Claude Sonnet 5.5 | 67.04 | #11 |
| 3 | Claude Opus 5.5 | 66.97 | #3 |
| 4 | Claude Fable 5.1 | 65.83 | #9 |
| 5 | Muse Wonder Full | 64.57 | #10 |
| 6 | Claude Opus 5 | 63.67 | #12 |
| 7 | GPT-6 Astra | 63.12 | #6 |
| 8 | Claude Fable 5 | 61.39 | #16 |
| 9 | GPT-6.1 Sol | 61.15 | #7 |
| 10 | Muse Wonder | 61.06 | #15 |
| 11 | Muse Spark 1.3 (Max) | 58.16 | #24 (Muse Spark 1.3) |
| 12 | GPT-5.6 Sol | 58.01 | #19 |
| 13 | GPT-6 Sol | 57.54 | #14 |
| 14 | DeepSeek V4.1 Flash (1017) | 56.58 | #20 |
| 15 | MiMo-V2.6-Pro | 55.2 | #42 |
Interactive version: theaggregate.ai/benchmark?slug=vals-ai-vals-index · How It Works · Data refreshed daily, snapshot 2026-10-20.