Kimi K3 needs ~1 TFLOPs per decoded token while V4.1F needs only 50 GFLOPs
YouJiacheng · x · 2026-10-26
You Jia Cheng compared decode-time compute requirements of Kimi K3 and K2 V4.1F: both models can be served in a compute-bound regime, but K3 needs 1 TFLOPs (FP8-equivalent) to decode one token versus only 50 GFLOPs for V4.1F — roughly a 20x difference.
He attributes the gap to implementation choices: V4.1F's MoE routed experts use W4A8 weights computed at MXFP8, with the rest in BF16. A useful concrete reference for anyone thinking about large-MoE inference cost and serving throughput.
More from Infra
- Anthropic S-1 reveals 48 cents of compute cost per dollar of Q2 revenue — nmasc_ · 2026-10-26
- Persistent AI joins UR network to route private AI inference with encrypted, IP-hidden traffic — 0xSammy · 2026-10-26
- Anthropic's disclosed compute commitments: $114.5B Google, $161.2B Broadcom top the list — BenBajarin · 2026-10-26
- SpaceX strikes $40B deal with Apollo Global to lease AI chips and data center hardware — Polymarket · 2026-10-26
- SpaceX AI revenue up 830%, Claude designs 318K antivenom proteins in 64 hours — PeterDiamandis · 2026-10-26
- Meta plans up to $95 billion stock sale to fund AI buildout — The Decoder · 2026-10-26