Kimi-K3
45 records across 5 hardware.
| hardware | model | quant | backend | tok/s | pp tok/s | W | ttft s | ctx/tg | date | class | flags | source |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MI355X | Kimi-K3 | BF16 | vLLM | 1083.0 | – | – | – | tg1024 | 2026-09-27 | community | – | source |
| MI355X | Kimi-K3 | BF16 | vLLM | 1082.6 | – | – | – | tg1024 | 2026-09-27 | community | – | source |
| GB300 | Kimi-K3 | NVFP4 | vLLM | 946 | – | – | – | 1024 | 2026-10-07 | sourced | – | source |
| GB300 | Kimi-K3 | NVFP4 | vLLM | 942 | – | – | – | 1024 | 2026-10-07 | sourced | – | source |
| GB300 | Kimi-K3 | NVFP4 | vLLM | 909 | – | – | – | 1024 | 2026-10-07 | sourced | – | source |
| GB300 | Kimi-K3 | NVFP4 | vLLM | 902 | – | – | – | 1024 | 2026-10-07 | sourced | – | source |
| 8x MI355X | Kimi-K3 | unspecified | vLLM | 693.13 | – | – | 336.53 | – | 2026-10-05 | sourced | – | source |
| 8x MI355X | Kimi-K3 | unspecified | vLLM | 685.62 | – | – | 340.08 | – | 2026-10-05 | sourced | – | source |
| MI355X | Kimi-K3 | BF16 | vLLM | 668.97 | – | – | – | tg1024 | 2026-09-28 | community | – | source |
| MI355X | Kimi-K3 | BF16 | vLLM | 658.42 | – | – | – | tg1024 | 2026-09-28 | community | – | source |
| MI355X | Kimi-K3 | MXFP4 | vLLM | 648 | – | – | – | 8192 | 2026-10-02 | community | – | source |
| MI355X | Kimi-K3 | MXFP4 | vLLM | 647 | – | – | – | 8192 | 2026-10-02 | community | – | source |
| MI355X | Kimi-K3 | MXFP4 | vLLM | 646 | – | – | – | 8192 | 2026-10-02 | community | – | source |
| MI355X | Kimi-K3 | MXFP4 | vLLM | 623 | – | – | – | 8192 | 2026-10-02 | community | – | source |
| MI355X | Kimi-K3 | FP4 | vLLM | 529.7 | – | – | – | 16384 | 2026-10-03 | community | – | source |
| MI355X | Kimi-K3 | FP4 | vLLM | 515.1 | – | – | – | 16384 | 2026-10-03 | community | – | source |
| MI355X | Kimi-K3 | FP4 | vLLM | 510.9 | – | – | – | 16384 | 2026-10-03 | community | – | source |
| MI355X | Kimi-K3 | FP4 | vLLM | 499.3 | – | – | – | 16384 | 2026-10-03 | community | – | source |
| MI355X | Kimi-K3 | FP4 | vLLM | 485.5 | – | – | – | 16384 | 2026-10-03 | community | – | source |
| MI355X | Kimi-K3 | FP4 | vLLM | 453.4 | – | – | – | 16384 | 2026-10-03 | community | – | source |
| GB300 | Kimi-K3 | NVFP4 | vLLM | 415 | – | – | – | 1024 | 2026-10-07 | sourced | – | source |
| GB300 | Kimi-K3 | NVFP4 | vLLM | 413 | – | – | – | 1024 | 2026-10-07 | sourced | – | source |
| MI355X | Kimi-K3 | BF16 | vLLM | 236.05 | – | – | – | tg1024 | 2026-09-27 | community | – | source |
| MI355X | Kimi-K3 | BF16 | vLLM | 236.05 | – | – | – | tg1024 | 2026-09-27 | community | – | source |
| MI355X | Kimi-K3 | unstated | vLLM | 233.11 | – | – | – | 1024 | 2026-10-13 | sourced | – | source |
| MI355X | Kimi-K3 | unstated | vLLM | 232.87 | – | – | – | 1024 | 2026-10-13 | sourced | – | source |
| MI355X | Kimi-K3 | unstated | vLLM | 190.9 | – | – | – | – | 2026-10-14 | sourced | – | source |
| MI355X | Kimi-K3 | unstated | vLLM | 180.98 | – | – | – | – | 2026-10-14 | sourced | – | source |
| MI355X | Kimi-K3 | unstated | vLLM | 179.77 | – | – | – | 1024 | 2026-10-13 | sourced | contradiction | source |
| MI355X | Kimi-K3 | unspecified | vLLM | 177.3 | – | – | – | 100000 | 2026-10-07 | community | – | source |
| MI355X | Kimi-K3 | unstated | vLLM | 175.7 | – | – | – | – | 2026-10-14 | sourced | – | source |
| MI355X | Kimi-K3 | unstated | vLLM | 174.25 | – | – | – | – | 2026-10-14 | sourced | – | source |
| MI355X | Kimi-K3 | unstated | vLLM | 172.89 | – | – | – | – | 2026-10-06 | sourced | – | source |
| MI355X | Kimi-K3 | unstated | vLLM | 164.79 | – | – | – | 1024 | 2026-10-13 | sourced | contradiction | source |
| MI355X | Kimi-K3 | unstated | vLLM | 163.94 | – | – | – | – | 2026-10-06 | sourced | – | source |
| 8x MI355X | Kimi-K3 | unstated | vLLM | 153.09 | – | – | 7291.12 | – | 2026-10 | sourced | – | source |
| 8x MI355X | Kimi-K3 | unstated | vLLM | 152.49 | – | – | 7280.8 | – | 2026-10 | sourced | – | source |
| 8x MI355X | Kimi-K3 | unstated | vLLM | 122.54 | – | – | 17312.48 | – | 2026-10 | sourced | – | source |
| 8x MI355X | Kimi-K3 | unstated | vLLM | 122.19 | – | – | 17272.42 | – | 2026-10 | sourced | – | source |
| GB300 | Kimi-K3 | NVFP4 | vLLM | 78.7 | – | – | – | 1024 | 2026-10-07 | sourced | – | source |
| GB300 | Kimi-K3 | NVFP4 | vLLM | 78.2 | – | – | – | 1024 | 2026-10-07 | sourced | – | source |
| GB300 | Kimi-K3 | NVFP4 | vLLM | 29.6 | – | – | – | 8192 | 2026-10-07 | sourced | – | source |
| GB300 | Kimi-K3 | NVFP4 | vLLM | 29.4 | – | – | – | 8192 | 2026-10-07 | sourced | – | source |
| 8×H100 (640 GB HBM) | Kimi-K3 | IQ1_S-XS | llama.cpp (CUDA) | 6.8 | – | – | – | 8192 | 2026-08-10 | community | – | source |
| M3 Ultra | Kimi-K3 | mxfp4 | oMLX | 6.5 | 95.8 | – | – | 4096 | 2026-10-17 | community | – | source |