Vals Index v2.2 (Vals Index)
We show this table for reference; we do not rank on it.
Vals AI GDP-weighted composite of eight finance, coding, legal, and tax benchmarks, updated to v2.2 on October 19, 2026.
Evaluation results published by Vals AI. Published leaderboard aggregates and evaluation settings only. Dataset contents are not included.
Vals Index score on Vals Index — October 27, 2026
We mirror the published vals index score view for Vals Index. Claude Sonnet 5.5 leads the public snapshot at 70.85%, followed by Gemini 4 Carbon (70.36%) and Claude Opus 5.5 (70.18%). We do not use these results to rank models overall.
Claude Sonnet 5.5
Anthropic
Gemini 4 Carbon
high reasoning
Claude Opus 5.5
Anthropic
29 modelsCross-domainCurrentDisplay onlyUpdated October 27, 2026
Vals Index score table (29 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Sonnet 5.5Anthropic | 70.85% | Not reported | Closed |
| 2 | Gemini 4 CarbonGooglehigh reasoning | 70.36% | Not reported | Closed |
| 3 | Claude Opus 5.5Anthropic | 70.18% | Not reported | Closed |
| 4 | Gemini 4 ArgonGooglehigh reasoning | 69.24% | Not reported | Closed |
| 5 | Muse Wonder FullMetamax reasoning | 66.79% | Not reported | Closed |
| 6 | GPT-6 AstraOpenAImax reasoning | 65.02% | Not reported | Closed |
| 7 | Muse WonderMetamax reasoning | 64.88% | Not reported | Closed |
| 8 | GPT-6.1 SolOpenAImax reasoning | 63.94% | Not reported | Closed |
| 9 | Kimi K3.1 ProMoonshot AImax reasoning | 62.52% | Not reported | Pending |
| 10 | DeepSeek V4.1 Flash 1017DeepSeekmax reasoning | 61.26% | Not reported | Open |
| 11 | GPT-6.1 TerraOpenAImax reasoning | 60.47% | Not reported | Closed |
| 12 | GLM-5.3Z.AImax reasoning | 59.73% | Not reported | Open |
| 13 | Qwen3.8 Max InternalAlibaba | 59.05% | Not reported | Unknown |
| 14 | Claude Haiku 5.5Anthropic | 58.73% | Not reported | Closed |
| 15 | Grok 4.7xAIxhigh reasoning | 58.56% | Not reported | Closed |
| 16 | MiMo-V2.6-ProXiaomi | 57.81% | Not reported | Open |
| 17 | Gemini 3.8 FlashGooglehigh reasoning | 55.95% | Not reported | Closed |
| 18 | MiMo-V2.6-FlashXiaomi | 54.83% | Not reported | Open |
| 19 | MiniMax M3.1 FlashMiniMax | 53.78% | Not reported | Closed |
| 20 | Hy4 previewTencent | 53.21% | Not reported | Open |
| 21 | DeepSeek V4 Pro 0813DeepSeekmax reasoning | 52.97% | Not reported | Open |
| 22 | Kimi K3Moonshot AImax reasoning | 52.95% | Not reported | Open |
| 23 | GPT-6 LunaOpenAImax reasoning | 52.32% | Not reported | Closed |
| 24 | GLM-5.3-FlashZ.AImax reasoning | 50.86% | Not reported | Open |
| 25 | Muse Spark OpenMetaxhigh reasoning | 48.57% | Not reported | Open |
| 26 | Mercury 3Inceptionhigh reasoning | 43.66% | Not reported | Closed |
| 27 | Beam 1.0Reflection | 43.42% | Not reported | Unknown |
| 28 | InklingThinking Machines Lab0.99 reasoning | 28.96% | Not reported | Open |
| 29 | Inkling-SmallThinking Machines Lab0.99 reasoning | 27.59% | Not reported | Open |
How to read this leaderboard
Operator receipt: 29 sourced rows are currently displayable on this page; the leading published row is Claude Sonnet 5.5 at 70.85%.
Honest limit: Component revisions prevent direct comparisons with earlier editions. Some Sonnet 5.5 and Opus 5.5 tasks use fallback models after provider refusals; counting those tasks as failures changes the published order.
How Vals Index is shown here
We mirror Vals Index v2.2, updated October 19, 2026. It combines eight benchmarks across finance, coding, legal, and tax work, with sector weights of 8.0, 5.6, 1.2, and 0.5 divided by 15.3.
Version 2.2 uses Finance Agent v2.1, EMB v1.1, Terminal-Bench 4.0, Vibe Code Bench v1.1, Code Migration v1.1, Legal Research, HLAB v1.1, and Tax Agent all-pass accuracy. Code Migration uses a fixed subset, so its index component differs from the standalone score.
Some Claude Sonnet 5.5 and Opus 5.5 tasks used fallback models after provider refusals. The published composite includes those results; treating them as failures changes the order. This composite stays display only and outside model rankings.
Snapshot
The published Vals Index snapshot places Claude Sonnet 5.5 first at 70.85%. The third row is 0.68 points behind. The broader top-10 range is 9.60 points, so many of the published results sit in a relatively narrow band.
29 models have been evaluated on Vals Index. The benchmark falls in the Cross-domain category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Vals Index is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Vals Index
Year
2026
Tasks
8 benchmarks across finance, coding, legal, and tax
Format
GDP-weighted sector means
Difficulty
Private economic-work benchmark composite
Version 2.2 uses Finance Agent v2.1, EMB v1.1, Terminal-Bench 4.0, Vibe Code Bench v1.1, Code Migration v1.1, Legal Research, HLAB v1.1, and Tax Agent all-pass accuracy. Sector means receive weights of 8.0 for finance, 5.6 for coding, 1.2 for legal, and 0.5 for tax, divided by 15.3. The Code Migration component uses a fixed subset rather than the full standalone set. We keep the composite display only.
Freshness and provenance
Version
Vals Index 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 Vals Index measure?
Vals AI GDP-weighted composite of eight finance, coding, legal, and tax benchmarks, updated to v2.2 on October 19, 2026.
Which model leads the published Vals Index snapshot?
Claude Sonnet 5.5 currently leads the published Vals Index snapshot with 70.85% vals index score. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Vals Index?
The October 27, 2026 snapshot contains 29 AI models.
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 6,100+ subscribers.
One email each week. Unsubscribe anytime.