Atlas

Benchmarks

← All benchmarks

Tax Agent Bench — Forms & Filings

Professional Work · 2026-09-23

Weighted rubric and citation credit on 30 tax forms-and-filings questions, using the parent evaluation protocol.

Top models (higher is better)

ModelScore
DeepSeek V4.1 Flash 101779.4
Muse Wonder Full76.6
MiniMax M3.1 Flash75.3
Claude Fable 5.173.2
GPT-5.6 Sol72.5
Grok 4.672.1
Muse Spark 1.371.7
GLM-5.371.7
Mistral Large 4 Preview71.7
Qwen3.8 Max (0902)71.5
Muse Wonder69.8
GPT-5.6 Terra68.7
Kimi K368.6
Hy4 Preview68.5
Gemini 4 Argon67.3
Claude Opus 567.2
Ember-165.9
Gemini 4 Carbon65.8
Sonnet 4.665.6
Sonnet 565.2
Gemini 3.8 Flash65.1
Grok 4.564.2
Claude Fable 564.0
Grok 4.763.8
GPT-6 Luna62.9
Qwen3.8 27B62.4
Kimi K3.1 Pro61.9
Claude Haiku 5.561.8
GLM-5.3 Flash61.4
GPT-6 Astra60.8
Opus 4.760.4
GPT-5.559.9
Muse Spark 1.259.5
DeepSeek-V4-Flash-073159.4
MiMo-V2.6-Pro59.1
Step 5 Preview58.8
GPT-5.6 Luna58.5
GPT-6.1 Sol57.7
DeepSeek V4.1 Flash56.8
Opus 4.855.8
Loading Atlas data…