Tencent Hunyuan internal bench scores its flagship 13.0% on hardest math tier
teortaxesTex · x · 2026-11-13
A rumor claims Chinese models may see a capability leap in mid-to-late December, sourced from a Hunyuan benchmark team member. Teortaxes considers the signal credible since Chinese models currently get thrashed on Hunyuan's hardest tasks.
Hunyuan's published internal table shows unusual transparency: its flagship Hy4 scores just 13.0% on Hy-FrontierMath T5 (internal), vs Claude Opus 5.5 at 97.5%, GPT 6.1 at 91.3%, and GLM 5.3 at 8.0%. Commentators note the wide remaining gap on deep symbolic math even as general benchmarks look close, and that the source isn't responsible for model training—so whether this is a general prediction or hype for the next Hunyuan remains to be seen by year-end.
More from Models
- Jason Calacanis threatens to ditch $20k/yr Claude as $40/mo Grok Bot outperforms it — chrisfleck · 2026-11-13
- Dan Shipper's weekend plan: let Claude Fable 5.2 build a custom Supernote OS — danshipper · 2026-11-13
- Alibaba teases Qwen4 with stronger coding and agentic capabilities at Nov 17 Hong Kong summit — teortaxesTex · 2026-11-13
- Baidu opens ERNIE 5.5 Flash Preview via OpenRouter — Baidu_Inc · 2026-11-13
- Hybrid Decision Model learns when to reason: 72.5% lower latency at higher accuracy — hbouammar · 2026-11-13
- AI×Bio benchmarks called out as LLM-generated "benchslop" that may be impairing bio-reasoning — anshulkundaje · 2026-11-13