Atlas

Benchmarks

← All benchmarks

Tax Agent Bench — All-Pass

Professional Work · 2026-09-23

Percentage of the 193 held-out questions passing every rubric item, without partial credit.

Top models (higher is better)

ModelScore
Claude Fable 5.149.2
Claude Opus 545.6
Gemini 4 Argon44.6
Muse Spark 1.243.5
Muse Spark 1.343.5
GLM-5.339.9
Claude Fable 538.3
Grok 4.734.2
Kimi K333.7
Grok 4.633.2
Hy4 Preview32.6
MiMo-V2.6-Pro32.6
Gemini 3.8 Flash32.1
GLM-5.3 Flash31.6
Ember-131.1
GPT-5.6 Sol31.1
Qwen3.8 27B31.1
MiMo-V2.6-Flash30.6
Mistral Large 4 Preview30.6
Opus 4.830.1
Sonnet 530.1
GPT-5.6 Terra30.1
DeepSeek-V4-Flash-073129.5
Step 5 Preview29.0
DeepSeek V4 Pro 081328.5
DeepSeek V4.1 Flash27.5
Sonnet 4.625.9
GPT-5.6 Luna24.9
Grok 4.524.9
Opus 4.723.8
Gemini 3.7 Flash22.8
Gemini 3.5 Flash21.8
GPT-5.521.8
GPT-6.1 Sol21.8
MiniMax M321.2
GPT-6 Astra20.7
GPT-6 Luna20.2
Inkling-Small18.7
Gemini 3.6 Flash18.1
Inkling17.6
Loading Atlas data…