Vals Index v2.2
GDP-weighted agentic performance across finance, coding, legal and tax work. Measured by Vals AI: Vals AI agentic harness.
measured 2026-11-18
Composite: (8.0 x Finance + 5.6 x Coding + 1.2 x Legal + 0.5 x Tax) / 15.3, each sector weighted by its share of US GDP. The broadest single measure here, and the closest thing to an overall ranking. Version 2.2 (2026-10-19) updated several component benchmarks and switched tax to all-pass accuracy, so its scores are not comparable with v2.1, and Vals has not yet rerun every model on it.
Vals Index v2.2 ranks 42 models; this table covers the 22 of them whose pricing we have verified against the provider's own page, so 20 scored models are not shown.
Vals AI — Vals Index v2.2, read 2026-11-20
Standings
| Model | Maker | Score | Blended $/MTok | Score per $ |
|---|---|---|---|---|
| 1Claude Fable 5.2 | Anthropic | 71.31% | $16.00 | 4.5 |
| 2Gemini 4 Neon | 70.87% | $6.00 | 11.8 | |
| 3Claude Sonnet 5.5 | Anthropic | 70.85% | $4.00 | 17.7 |
| 4Gemini 4 Carbon | 70.45% | $4.00 | 17.6 | |
| 5Claude Opus 5.5 | Anthropic | 70.14% | $8.00 | 8.8 |
| 6Gemini 4 Argon | 69.39% | $4.00 | 17.3 | |
| 7Gemini 4 Hydrogen | 68.49% | $3.00 | 22.8 | |
| 8gpt-6.2-astra | OpenAI | 67.66% | $16.00 | 4.2 |
| 9gpt-6.2-sol | OpenAI | 66.06% | $4.00 | 16.5 |
| 10gpt-6-astra | OpenAI | 65.02% | $20.00 | 3.3 |
| 11gpt-6.1-sol | OpenAI | 63.94% | $4.00 | 16.0 |
| 12Kimi K3.1 Pro | Moonshot | 62.52% | $4.00 | 15.6 |
| 13DeepSeek Flash V4.1 Flash 1017 | DeepSeek | 61.26% | $0.525 | 116.7 |
| 14gpt-6.1-terra | OpenAI | 60.54% | $2.00 | 30.3 |
| 15GLM-5.3 | Z.ai | 59.85% | $2.15 | 27.8 |
| 16Claude Haiku 5.5 | Anthropic | 58.73% | $0.2 | 293.6 |
| 17Grok 4.7 | xAI | 58.62% | $3.00 | 19.5 |
| 18Gemini 3.8 Flash | 55.95% | $1.50 | 37.3 | |
| 19DeepSeek V4 Pro 0813 | DeepSeek | 52.97% | $1.98 | 26.8 |
| 20Kimi K3 | Moonshot | 52.95% | $6.00 | 8.8 |
| 21gpt-6-luna | OpenAI | 52.32% | $0.2 | 261.6 |
| 22GLM-5.3-Flash | Z.ai | 50.86% | $0.237 | 214.1 |