community pool / model answer

Kimi-Linear-48B-A3B-Instruct

chat · moonshotai/Kimi-Linear-48B-A3B-Instruct

233.1 tok/s · model median · 6 runs in the model record.

Reference rigQuantizationMedian resultBasisTTFT (ms)Peak VRAMMax contextEngines
RTX PRO 6000 Blackwell 96GB5-bit267 tok/sreported75- GB262,144llama.cpp
RTX PRO 6000 Blackwell 96GB6-bit256.3 tok/sreported107- GB262,144llama.cpp
RTX PRO 6000 Blackwell 96GB8-bit119 tok/smeasured416,256- GB1,000,000llama.cpp

Context tested

The table above is the tested context for each exact rig and quantization cell. Empty fields remain empty rather than being filled with an estimate.