Hy-4 vs DeepSeek Flash benchmark: quarter the tokens, a third of requests
ShunyuYao12 · x · 2026-11-05
Developer loktar00 benchmarked Hy-4 against DeepSeek Flash using the Deepseek Harness. He still prefers DeepSeek's output quality, but Hy4 put up a strong showing with about a quarter of the tokens and a third of the requests.
More from Models
- Why Opus 5 speaks Claudist: RL-side hypotheses for Claude's dialect — burny_tech · 2026-11-05
- Racing all 4 Claude models in a sailing simulator to verify the pecking order — shekitup · 2026-11-05
- Tencent open-sources Hy4 preview: 770B params, 49B active, 1M-token context — emmanuelvivier · 2026-11-05
- Moonshot AI open-sources Kimi K3.1 Pro weights, a 2.8T-parameter model — emmanuelvivier · 2026-11-05
- Llama app adds one-click local decision models: d3-flash tops Jev Index, d1-Omni-2B goes omni — huggingface · 2026-11-05
- SII-GAIR applies RSI to full frontier-scale pretraining: 30B model trained on 134B agentic tokens — AdinaYakup · 2026-11-05