Models › Kimi-Linear-48B-A3B-Instruct

Kimi-Linear-48B-A3B-Instruct: VRAM needs and speed by GPU

At Q4_K_M, Kimi-Linear-48B-A3B-Instruct needs about 32.1 GB of VRAM. No measured runs yet. Speeds below are estimates. Submit a run.

49.12B parameters (mixture-of-experts, ~3.1B active) · license: mit · 1,048,576-token context · GGUF published Feb 6, 2026 · GGUF files on Hugging Face

Memory needed per quantization

QuantFile sizeNeeds (estimate)Smallest card it fits
IQ1_S10.5 GB12.5 GB16 GB cards
IQ1_M10.9 GB13.0 GB16 GB cards
IQ2_XXS12.1 GB14.2 GB16 GB cards
IQ2_XS13.9 GB15.9 GB24 GB cards
IQ2_S14.0 GB16.0 GB24 GB cards
IQ2_M15.8 GB17.8 GB24 GB cards
Q2_K17.5 GB19.6 GB24 GB cards
Q2_K_L17.9 GB19.9 GB24 GB cards
IQ3_XXS19.7 GB21.7 GB24 GB cards
IQ3_XS20.4 GB22.5 GB24 GB cards
Q3_K_S21.6 GB23.6 GB32 GB cards
IQ3_M22.7 GB24.7 GB32 GB cards
Q3_K_M22.7 GB24.7 GB32 GB cards
Q3_K_L23.6 GB25.6 GB32 GB cards
Q3_K_XL23.9 GB26.0 GB32 GB cards
IQ4_XS26.5 GB28.5 GB32 GB cards
IQ4_NL28.0 GB30.0 GB32 GB cards
Q4_028.4 GB30.5 GB32 GB cards
Q4_K_S29.0 GB31.0 GB32 GB cards
Q4_K_M30.1 GB32.1 GB48 GB cards
Q4_K_L30.3 GB32.4 GB48 GB cards
Q4_131.0 GB33.0 GB48 GB cards
Q5_K_S34.0 GB36.0 GB48 GB cards
Q5_K_M35.1 GB37.1 GB48 GB cards
Q5_K_L35.3 GB37.4 GB48 GB cards
Q6_K40.5 GB42.5 GB48 GB cards
Q6_K_L40.6 GB42.7 GB48 GB cards
Q8_052.2 GB54.3 GB96 GB cards
BF1698.3 GB100.3 GBmulti-GPU / large unified memory

Expected speed on popular hardware

HardwareBest quant that fitsExpected tok/s
NVIDIA RTX 3060Does not fit
NVIDIA RTX 4060 TiDoes not fit
NVIDIA RTX 3090IQ3_XXS267 estimate
NVIDIA RTX 4090IQ3_XXS296 estimate
NVIDIA RTX 5090Q4_K_S317 estimate
AMD Radeon RX 7900 XTXIQ3_XXS297 estimate
Apple M4 Pro (20-core GPU) (64 GB)Q6_K45.8 estimate
Apple M3 Max (40-core GPU) (64 GB)Q6_K66.9 estimate
AMD Ryzen AI Max+ 395 (Radeon 8060S)Q8_031.8 estimate

Check your own card

Too big for your card?

Run Kimi-Linear-48B-A3B-Instruct on a rented 24-80 GB GPU by the hour.

Rent a bigger GPU on RunPod

Affiliate link: if you sign up, RigBench may earn a referral credit. It never affects rankings or numbers.