Hy4 Official Release EvaluationProvider report
2026-11-07Hunyuan 4 (Hy4) only: 34 metrics from the official release appendix plus 3 items from the post body (finance expert crowdtest and self-built search tests), 37 in total. All are vendor-reported results at the highest available reasoning setting; benchmarks marked * in the table were first evaluated at release and are flagged in raw. DeepSWE v1.1 uses mini-swe-agent; ProgramBench records the Almost Solved metric; the Hy4 HLE-Full (with tools) value carries no text-only marker and is treated as the full set; the Agents' Last Exam v1.1 and ZeroBench-main tooling and subsets differ from the Kimi K3.1 Pro report and are stored separately. Comparison-model columns (GLM-5.3, DeepSeek-V4.1-Flash-1017, Kimi K3.1 Pro, GPT-6.1 Sol, Claude Opus 5.5) are not imported. The vendor notes a remaining gap to top closed models and plans to open-source weights and the system card after release.