Atlas

Benchmarks

← All benchmarks

Terminal-Bench 4.0 — Software

Agents · 2026-09-22

Resolution rate on the 18 software tasks in Terminal-Bench 4.0.

Top models (higher is better)

ModelScore
Gemini 4 Carbon63.0
GPT-6 Astra61.1
GPT-6.1 Sol57.4
Muse Wonder57.4
GPT-6 Sol53.7
Kimi K3.1 Pro53.7
Muse Wonder Full53.7
GPT-6.1 Terra50.0
Gemini 4 Argon48.1
Claude Haiku 5.546.3
GPT-5.6 Sol46.3
Qwen3.8 Max (0902)44.4
MiMo-V2.6-Pro38.9
DeepSeek V4.1 Flash 101737.0
GLM-5.3 Flash35.2
GLM-5.335.2
GPT-5.6 Terra35.2
Mistral Large 4 Preview35.2
Step 5 Preview35.2
DeepSeek V4.1 Flash33.3
Grok 4.729.6
MiMo-V2.6-Flash29.6
Muse Spark 1.329.6
Ember-125.9
Grok 4.625.9
DeepSeek-V4-Flash-073124.1
DeepSeek V4 Pro 081322.2
Kimi K322.2
MiniMax M3.1 Flash22.2
Gemini 3.8 Flash18.5
Opus 4.816.7
Sonnet 514.8
GPT-5.6 Luna14.8
GPT-6 Luna14.8
Hy4 Preview14.8
Beam 1.0 501B A23B11.1
Muse Spark 1.211.1
Muse Spark Open11.1
Gemini 3.7 Flash7.4
GPT-5.4 Mini7.4
Loading Atlas data…