Vals AI TaxEval v2: leaderboard
One item is a hard tax question written and double-checked by tax experts, of types such as compliance, comparative analysis, numerical reasoning, semantic analysis and current updates. More than 1,500 questions sit mostly in a private test set, and a judge model grades answer correctness and stepwise reasoning as two separate tasks. This board is the overall accuracy, the average of the two, higher is better.
Metric: Accuracy (%). Source: www.vals.ai. Saturation forecast: Around March 2027. 145 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Muse Spark 1.2 (xHigh) | 80.38 | #62 (Muse Spark 1.2) |
| 2 | Muse Spark 1.1 (xHigh) | 79.72 | #52 (Muse Spark 1.1) |
| 3 | Muse Spark | 77.68 | #64 |
| 4 | Claude Sonnet 4.6 (Max) | 77.11 | #90 (Claude Sonnet 4.6) |
| 5 | Claude Fable 5 (Max) | 76.94 | #16 (Claude Fable 5) |
| 6 | GPT-5.6 Luna (Max) | 76.17 | #68 (GPT-5.6 Luna) |
| 7 | GPT-5.6 Terra (xHigh) | 76.17 | #41 (GPT-5.6 Terra) |
| 8 | Claude Fable 5.1 (Max) | 75.96 | #9 (Claude Fable 5.1) |
| 9 | Claude Opus 4.6 (Adaptive Reasoning, Max Effort) | 75.96 | #63 (Claude Opus 4.6) |
| 10 | Grok 3 | 75.88 | #304 |
| 11 | GPT-5.2 (2025-12-11) (xHigh) | 75.76 | |
| 12 | Kimi K3 | 75.72 | #30 |
| 13 | Grok 4 Fast (Reasoning) | 75.7 | #254 (Grok 4 Fast) |
| 14 | Claude Opus 4.8 (Max) | 75.63 | #37 (Claude Opus 4.8) |
| 15 | Claude Sonnet 5 (Max) | 75.63 | #57 (Claude Sonnet 5) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=vals-ai-taxeval-v2 · How It Works · Data refreshed daily, snapshot 2026-10-20.