Atlas

Benchmarks

← All benchmarks

Code Migration (Vals Index v2 subset)

Code · 2026-08-13

Vals Index v2's Code Migration score uses a fixed subset of 50 of the standalone board's 120 CLI tasks plus all 10 COBOL tasks, weighted 75% CLI and 25% COBOL. The subset was selected to approximate full-board rankings; its score is not relabeled as the full evaluation.

Top models (higher is better)

ModelScore
Gemini 4 Argon69.5
GPT-6 Astra68.5
GPT-6.1 Sol63.3
Muse Wonder62.0
Muse Wonder Full60.3
GPT-6 Sol56.9
Claude Opus 556.7
GPT-5.6 Sol52.4
DeepSeek V4.1 Flash 101752.2
Muse Spark 1.348.3
GPT-5.6 Terra48.2
Opus 4.846.9
DeepSeek V4.1 Flash46.9
Hy4 Preview46.8
Claude Haiku 5.545.9
GPT-5.6 Luna44.8
Grok 4.644.6
GPT-6 Luna44.4
Sonnet 544.1
GLM-5.343.2
Grok 4.743.2
DeepSeek-V4-Flash-073142.3
MiMo-V2.6-Pro42.3
DeepSeek V4 Pro 081341.5
MiMo-V2.6-Flash40.4
Step 5 Preview39.5
Grok 4.538.4
Gemini 3.8 Flash31.6
Ember-130.5
Muse Spark 1.229.4
Gemini 3.7 Flash29.0
Gemini 3.5 Flash27.1
DeepSeek-V4-Pro26.2
Qwen3.8-Max23.2
Mistral Large 4 Preview22.8
Muse Spark Open21.2
MiniMax M320.9
MiMo-V2.5-Pro20.4
GLM-5.3 Flash20.2
Gemini 3.1 Pro Preview17.9
Loading Atlas data…