Sign in

Tax Agent Bench

As of Nov 3, 2026, Tax Agent Bench (v1) is maintained by Vals AI. Metric accuracy (%). Ceiling 100. Saturation unknown. Contamination risk low.

Maintainer
Vals AI
Metric
accuracy (%)
Ceiling
100
Saturation
unknown
Contamination
low
Cadence
Declared cadence: as Vals updates (checked daily). Catalog snapshots: 1, latest Nov 2, 2026.
Source
Vals AI · this benchmark on Vals AI
Grading
AI judge (LLM-graded rubric) — Research-grade corporate tax questions graded by an LLM judge against expert rubrics.

Source: Vals AI. A proprietary private test set run by Vals; scores are republished with permission. Rows are dated the day isitsota first saw them, and only the benchmark's standard harness counts.

Categories

Where it counts

Top 5

Current top scores

  1. Gemini 4 Carbon · 51.785 · Nov 2, 2026
  2. Claude Fable 5.1 · 49.194 · Nov 2, 2026
  3. Claude Opus 5.5 · 45.573 · Nov 2, 2026
  4. Claude Opus 5 · 45.548 · Nov 2, 2026
  5. Gemini 4 Argon · 44.56 · Nov 2, 2026
ModelScoreDateKind
Claude Fable 5.149.1942026-11-02snapshot
Claude Opus 5.545.5732026-11-02snapshot
Claude Opus 545.5482026-11-02snapshot
Gemini 4 Argon44.562026-11-02snapshot
DeepSeek V4.1 Flash 101744.352026-11-02snapshot
Muse Wonder Full42.9972026-11-02snapshot
Muse Spark 1.342.4352026-11-02snapshot
Claude Sonnet 5.541.9592026-11-02snapshot
GLM 5.339.6982026-11-02snapshot
Claude Fable 538.2812026-11-02snapshot
Muse Wonder36.2692026-11-02snapshot
Muse Spark 1.335.9742026-11-02snapshot
Grok 4.734.1972026-11-02snapshot
Kimi K333.3682026-11-02snapshot
Grok 4.633.1412026-11-02snapshot
Muse Spark 1.232.7982026-11-02snapshot
Hy4 Preview32.6172026-11-02snapshot
Mimo V2.6 Pro32.3882026-11-02snapshot
MiniMax M3.1 Flash32.1362026-11-02snapshot
Gemini 3.8 Flash32.1242026-11-02snapshot
Qwen3 8 Max32.0952026-11-02snapshot
Claude Haiku 5.531.9222026-11-02snapshot
GLM 5.3 Flash31.552026-11-02snapshot
GPT-5.6 Sol31.0882026-11-02snapshot
Qwen3.8 27B30.8942026-11-02snapshot
Mistral Large 430.3912026-11-02snapshot
GPT-5.6 Terra30.0352026-11-02snapshot
Claude Opus 4.829.9992026-11-02snapshot
Claude Sonnet 529.3742026-11-02snapshot
Mimo V2.6 Flash28.5072026-11-02snapshot