Tencent open-sources 770B MoE model with 49B active params, 1M context and a 214GiB quantized path
future_coded · x · 2026-09-05
Tencent Hunyuan released an open-source Apache 2.0 text model: 770B total params, 49B active per token, native 1M context, with Day 0 vLLM/SGLang support and an official FP8 checkpoint.
Quantization route
- Mixed GGUF compresses weights from 1.5TB to 214GiB with reported accuracy close to BF16
- It's per-layer mixed quantization, not a blanket 1-bit model: sensitive layers keep more precision
Reality check (preview)
- GGUF requires a patched llama.cpp path for hyv4
- Offload is still offload — not a laptop-native 770B
- Measure tokens/s on your own infra and compare BF16/FP8 vs mixed GGUF on real tasks
The story is ownership: download the weights, serve them yourself, and keep lowering the deployment bar without losing the 49B-active capability.
Related event: Tencent Open-Sources Hunyuan Hy4 Preview with 770B Params and 1M Context(3 posts)→
More from Infra
- FP16 vs FP32 matmul on RTX 4060: 3.5x speedup at large sizes, slower at 128x128 — cneuralnetwork · 2026-10-23
- BaseQuant V1 quants for agentic Qwen3.8-27B hit 17-34% lower KL than Unsloth — nightowlsleeping · 2026-10-23
- PhD students in some countries get B200s on demand, unlike their US peers — HildeKuehne · 2026-10-23
- Nvidia's Berlin GTC spotlights humanoid robots and Vera Rubin AI supercomputing — emmanuelvivier · 2026-10-23
- Lenovo adds 5 AI Library use cases, claims 1-week deployment and up to 70% lower token cost — shashib · 2026-10-23
- Linux Kernel Module Unlocks Half-Speed Tensor Cores on RTX 3090, Boosting BF16 to 1.9x — UltramanQuar · 2026-10-23