Atlas

Benchmarks

← All benchmarks

IOI 2026 (Vals v2)

Code

The six IOI 2026 problems under Vals' v2 OpenCode protocol without interactive grading feedback. Final solutions receive official subtask points, averaged across the problems.

Top models (higher is better)

ModelScore
GPT-5.6 Terra100.0
GPT-6 Astra100.0
Claude Fable 5.185.2
Claude Opus 5.585.2
GPT-5.6 Sol85.2
GPT-6 Sol68.5
Claude Opus 565.8
Gemini 3.7 Flash60.8
Qwen3.8-Max60.2
GPT-6 Luna58.3
Kimi K357.8
DeepSeek V4 Pro 081356.3
Grok 4.755.3
Hy4 Preview55.3
Gemini 3.8 Flash53.8
GLM-5.353.5
Muse Spark 1.352.5
GPT-5.6 Luna50.5
GLM-5.3 Flash47.7
Grok 4.643.5
DeepSeek V4.1 Flash43.0
Sonnet 542.5
GPT-5.3-Codex41.7
Grok 4.541.0
Qwen3.8 27B36.8
DeepSeek-V4-Flash-073136.0
MiMo-V2.6-Pro35.2
Gemini 3.6 Flash23.7
Muse Spark 1.220.2
Inkling19.7
Inkling-Small1.7
Mercury 2.51.2
Loading Atlas data…