Models › Kolibri-1

Kolibri-1: VRAM needs and speed by GPU

At Q4_K_M, Kolibri-1 needs about 52.5 GB of VRAM. No measured runs yet. Speeds below are estimates. Submit a run.

78.1B parameters · license: apache-2.0 · 262,144-token context · GGUF published Oct 18, 2026 · GGUF files on Hugging Face

Memory needed per quantization

QuantFile sizeNeeds (estimate)Smallest card it fits
IQ1_S17.6 GB20.4 GB24 GB cards
IQ1_M19.8 GB22.5 GB24 GB cards
IQ2_XXS22.8 GB25.5 GB32 GB cards
IQ2_XS24.3 GB27.1 GB32 GB cards
IQ2_S25.6 GB28.4 GB32 GB cards
IQ2_M28.1 GB30.8 GB32 GB cards
Q2_K29.9 GB32.7 GB48 GB cards
IQ3_XXS33.7 GB36.4 GB48 GB cards
IQ3_XS35.5 GB38.2 GB48 GB cards
Q3_K_S35.9 GB38.7 GB48 GB cards
Q3_K_M37.6 GB40.4 GB48 GB cards
Q3_K_L39.8 GB42.6 GB48 GB cards
IQ3_M40.7 GB43.4 GB48 GB cards
IQ4_XS43.6 GB46.4 GB48 GB cards
Q4_044.6 GB47.4 GB48 GB cards
Q4_K_S46.4 GB49.2 GB96 GB cards
Q4_149.2 GB52.0 GB96 GB cards
IQ4_NL49.8 GB52.5 GB96 GB cards
Q4_K_M49.8 GB52.5 GB96 GB cards
Q4_K_L53.5 GB56.3 GB96 GB cards
Q5_K_S55.8 GB58.5 GB96 GB cards
Q5_K_M60.3 GB63.0 GB96 GB cards
Q6_K_S65.8 GB68.6 GB96 GB cards
Q6_K69.0 GB71.7 GB96 GB cards
Q6_K_L72.5 GB75.3 GB96 GB cards
Q8_083.1 GB85.9 GB96 GB cards
BF16156.3 GB159.1 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 3090IQ1_M33.2 estimate
NVIDIA RTX 4090IQ1_M36.7 estimate
NVIDIA RTX 5090IQ2_M40.9 estimate
AMD Radeon RX 7900 XTXIQ1_M36.9 estimate
Apple M4 Pro (20-core GPU) (64 GB)IQ4_XS5.3 estimate
Apple M3 Max (40-core GPU) (64 GB)IQ4_XS7.7 estimate
AMD Ryzen AI Max+ 395 (Radeon 8060S)Q8_02.5 estimate

Check your own card

Too big for your card?

Run Kolibri-1 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.