antirez: Kimi K3.1 runs locally at 1 token/sec, good enough for interactive QA
antirez · x · 2026-11-08
antirez (creator of Redis) confirms that Kimi K3.1 PRO can be inferred locally via DwarfStar's kimi-k3 branch: generation runs at about 1 token/sec — usable interactively — while prefill is much faster, especially for long questions.
Related event: antirez's DwarfStar runs Kimi K3.1 PRO locally from safetensors(4 posts)→
More from Infra
- Anthropic to pay $3.4B for gas-fired generators, largest power gear purchase by an AI lab — GaryMarcus · 2026-11-08
- Meta, Google and a dozen more giants quietly bought Texas land for data centers — csuwildcat · 2026-11-08
- antirez: most local inference is single-user, just attach pi or opencode to local server — antirez · 2026-11-08
- antirez: large batches switch to layer-major inference, nearly as fast as full-resident — antirez · 2026-11-08
- Vitalik backs anti-data-center protests: algorithmic progress makes bigger compute unnecessary — paulnovosad · 2026-11-08
- AI costs falling faster than any past tech, says analyst chart — oran_ge · 2026-11-08