Tencent Hunyuan preview: 770B params, 1M context, Apache 2.0, 214GiB quantized
Aiden_Tech_Ai · x · 2026-09-05
The AI model race is entering a new phase: beyond "whose model is smartest," the question is now "who lets builders own and operate it." Tencent's Hunyuan preview is a strong example.
Key specs:
- 770B total parameters, 49B active per token
- Native 1M context window
- Apache 2.0 license
Mixed per-layer quantization shrinks the footprint from 1.5TB to 214GiB, with Tencent reporting only minor accuracy loss versus BF16. Still heavy, but "heavy and deployable" is very different from "open in name only."
More from Infra
- Dev boosts GLM 5.2 TPS on a B300 and swaps it into Claude Code in place of Anthropic models — abhijithneil · 2026-09-05
- What forces LLM teams to optimize inference when going from MVP to production? — Ok_Philosophy_4031 · 2026-09-05
- Leaker claims GPT-6 Astra beats Claude Fable 5.1 at metaprompting — whoiskatrin · 2026-09-05
- Local Qwen 27B vibecodes a playable Godot dungeon game in just 4 prompts — jacek2023 · 2026-09-05
- Open-sourcing nopasswd-sudo: time-boxed passwordless sudo for agents, built by a local 124B model — max_paperclips · 2026-09-05
- Experiential Labs: open-source gateway unifying 1,000+ AI models with one key and zero markup — Shruti_0810 · 2026-09-05