Atlas

Benchmarks

← All benchmarks

SAGE v1.1 - Linear Algebra

Math · 2026-10-27

The Linear Algebra subject of SAGE v1.1, scored by class-balanced check accuracy over rubric deductions on handwritten linear algebra solutions. Vals notes the subject has only 27 applied deductions, so its scores are noisy.

Top models (higher is better)

ModelScore
GLM-5.3 Flash87.7
Claude Opus 5.587.5
GPT-6.1 Sol87.2
Claude Haiku 5.586.8
Gemini 3.8 Flash84.0
Qwen3.7-Plus (source unspecified)83.1
Grok 4.783.1
Qwen3.5-Flash (source unspecified)83.1
Gemini 4 Argon81.9
Claude Sonnet 5.580.3
MiMo-V2.6-Flash80.2
Qwen3.5 397B A17B79.6
GPT-5.6 Terra79.5
Kimi K379.1
MiniMax M378.9
MiMo-V2.6-Pro78.0
Ember-177.1
Muse Spark Open76.7
GPT-6 Luna76.5
MiniMax M3.1 Flash74.7
DeepSeek V4.1 Flash72.7
Gemma 4 31B IT72.3
Gemini 3.5 Flash-Lite70.0
Gemini 3.1 Pro Preview67.2
Mistral Medium 3.564.8
Grok 4.1 Fast64.2
Mistral Small 459.7
Loading Atlas data…