Atlas

Benchmarks

← All benchmarks

Code Migration v1.1 - CLI

Code · 2026-10-19

The CLI split of Code Migration v1.1: 30 open-source Python, Java, Kotlin, Rust and C++ repositories, each migrated into its four non-source languages on the Pi harness and graded by hidden command-line behavior tests. A repository's score averages its four target languages.

Top models (higher is better)

ModelScore
Claude Sonnet 5.582.8
GPT-6 Astra76.4
GPT-6.1 Sol74.8
Gemini 4 Argon67.8
Muse Wonder67.1
DeepSeek V4.1 Flash 101759.2
Muse Wonder Full52.7
DeepSeek V4.1 Flash47.7
GPT-6 Luna44.3
Claude Haiku 5.542.1
GLM-5.340.6
Grok 4.740.6
MiMo-V2.6-Pro39.2
GLM-5.3 Flash38.7
Hy4 Preview37.9
DeepSeek V4 Pro 081336.8
Step 5 Preview35.2
Mercury 328.0
Gemini 3.8 Flash27.6
MiMo-V2.6-Flash27.0
Muse Spark Open15.0
Kimi K311.4
Beam 1.0 501B A23B8.9
MiniMax M38.3
Inkling-Small3.0
Inkling2.6
Loading Atlas data…