Hy4 preview

腾讯 Hy4 官方发布比较表 · 官方发布比较 · Diamond

腾讯官方比较表 · Hy4 preview · GPQA Diamond; highest available reasoning; starred Tencent runs

huggingface.co · 官方发布比较 · current published task suite

Hy4 preview / official APEX-Agents pass@1 · Original numeric labels visually read in official image; APEX-Agents; Official Hy4 preview blue series; not Hy3 cyan sub-series. JPEG read at decoder 1/8 scale, original SHA-256 preserved.

Vals AI · 独立评测 · 4.0

Hy4 preview / Mini-SWE-agent · Current displayed evaluator row. Mini-SWE-agent. Scores are not pooled across harnesses; source methodology and repeat counts retained.

腾讯 Hy4 官方发布比较表 · 官方发布比较 · current published task suite

腾讯官方比较表 · Hy4 preview · AutomationBench v1.0.6; highest available reasoning; starred Tencent runs

Vals AI · 独立评测 · 1.1

Hy4 preview / Published Code Migration 1.1 sandbox · Current displayed evaluator row. Published Code Migration 1.1 sandbox. Scores are not pooled across harnesses; source methodology and repeat counts retained.

腾讯 Hy4 官方发布比较表 · 官方发布比较 · 1.1

腾讯官方比较表 · Hy4 preview · DeepSWE current 1.1 task suite; mini-swe-agent; sandbox 8 CPUs and 16 GB RAM