Kimi-K3 weights land on Hugging Face, with GGUF builds ready for llama.cpp
MaziyarPanahi · x · 2026-08-04
Kimi-K3 weights are now on Hugging Face, with GGUF builds for local inference
The post says moonshotai/Kimi-K3 weights are available on Hugging Face, and that Unsloth already provides GGUF builds.
- Quantizations range from UD-IQ1S up to UD-Q8KXL.
- The weights can be run locally with llama.cpp.
- Users are told to choose a quantization level that fits their RAM.
- The screenshot shows the Hugging Face model card for unsloth/Kimi-K3-GGUF.
- The reply also mentions a “Jellyfish eval” where Kimi K3 streamed 6,522 characters of canvas code.
More from Models
- Epoch AI’s MirrorCode benchmark sees Claude Fable solve C preprocessor and Pkl tasks — Jsevillamol · 2026-08-04
- X post jokes that Anthropic could hit $100B in revenue by year end — Jsevillamol · 2026-08-04
- DeepSeek chatter says the company is moving beyond a V4-Ultra-style model — teortaxesTex · 2026-08-04
- Windows reset restores RTX 3070 throughput for local Qwen3.6-35B inference — campaigner_ · 2026-08-04
- ChatGPT’s anti-sycophancy training may be causing “performative nuance” — Chance-Physics-7216 · 2026-08-04
- Qwen 3.8 Max scores well on analysis tasks, Reddit says it is a cheap Opus 4.7 replacement — Ill_Distribution8517 · 2026-08-04