Atlas

Benchmarks

← All benchmarks

SWE-bench Multimodal (Vals v1): Scratch GUI

Code

Vals v1 issue-resolution rate on its Scratch GUI repository panel; one offline mini-swe-agent run with screenshot attachments and official test grading. Subset size is not published.

Top models (higher is better)

ModelScore
Claude Opus 5.550.0
Claude Sonnet 5.550.0
Gemini 4 Argon50.0
Claude Fable 5.133.3
Muse Wonder Full33.3
Muse Wonder33.3
Claude Haiku 5.516.7
DeepSeek V4.1 Flash16.7
Gemini 3.8 Flash16.7
GPT-6.1 Sol16.7
GPT-6 Astra16.7
GPT-6 Luna16.7
Grok 4.716.7
Kimi K316.7
MiMo-V2.6-Pro16.7
Step 5 Preview16.7
MiniMax M30.0
Mistral Large 4 Preview0.0
Muse Spark 1.30.0
Loading Atlas data…