// Open-model VRAM table · family
Kimi: local deployment & VRAM
14 open-weight models from Moonshot AI (17 official repos incl. quantised builds). At 4-bit they range from 11.4 GB (Moonlight-16B-A3B-Instruct) to 1.87 TB (Kimi-K3).
official orgs: moonshotailatest 2026-06-13memory = estimate
01Text LLM (6)
| Model | Params | Context | BF16 | 4-bit | Smallest GPU (4-bit) | Mac (4-bit) | Licence | Released |
|---|---|---|---|---|---|---|---|---|
| Kimi-K2-Thinking | 1T / 32B | 256K | 2.20 TB | 668 GB | 8 × H200 (1,128GB) | — | Commercial with conditions | 2025-11-04 |
| Kimi-Linear-48B-A3B-Instruct | 49.1B / 3B | — | 109 GB | 33.3 GB | 2 × RTX 5090 (64GB) | Mac 64GB | Commercial OK | 2025-10-30 |
| Kimi-K2-Instruct-0905 | 1T / 32B | 256K | 2.20 TB | 668 GB | 8 × H200 (1,128GB) | — | Commercial with conditions | 2025-09-03 |
| Kimi-K2-Instruct | 1T / 32B | 128K | 2.20 TB | 668 GB | 8 × H200 (1,128GB) | — | Commercial with conditions | 2025-07-11 |
| Kimi-Dev-72B ↗ | 72.7B | 128K | 163 GB | 51.7 GB | 2 × RTX 5090 (64GB) | Mac 128GB | Commercial OK | 2025-06-16 |
| Moonlight-16B-A3B-Instruct | 16B / 3B | 8K | 35.9 GB | 11.4 GB | 16GB GPU | Mac 24GB | Commercial OK | 2025-02-22 |
02Vision-language (7)
| Model | Params | Context | BF16 | 4-bit | Smallest GPU (4-bit) | Mac (4-bit) | Licence | Released |
|---|---|---|---|---|---|---|---|---|
| Kimi-K3 | 2.8T / 104B | 1M | 6.16 TB | 1.87 TB | 8 × B300 (2,304GB) | — | Custom licence – read it | 2026-06-13 |
| Kimi-K2.7-Code | 1T / 32B | 256K | 2.20 TB | 668 GB | 8 × H200 (1,128GB) | — | Commercial with conditions | 2026-06-11 |
| Kimi-K2.6 | 1T / 32B | 256K | 2.20 TB | 668 GB | 8 × H200 (1,128GB) | — | Commercial with conditions | 2026-04-14 |
| Kimi-K2.5 | 1T / 32B | 256K | 2.20 TB | 668 GB | 8 × H200 (1,128GB) | — | Commercial with conditions | 2026-01-01 |
| Kimi-VL-A3B-Thinking-2506 | 16.4B / 3B | 128K | 36.8 GB | 11.7 GB | 16GB GPU | Mac 24GB | Commercial OK | 2025-06-21 |
| Kimi-VL-A3B-Instruct | 16.4B / 3B | 128K | 36.8 GB | 11.7 GB | 16GB GPU | Mac 24GB | Commercial OK | 2025-04-09 |
| Kimi-VL-A3B-Thinking | 16.4B / 3B | 128K | 36.8 GB | 11.7 GB | 16GB GPU | Mac 24GB | Commercial OK | 2025-04-09 |
03Text-to-speech / audio (1)
| Model | Params | BF16 | 4-bit | Smallest GPU (4-bit) | Mac (4-bit) | Licence | Released |
|---|---|---|---|---|---|---|---|
| Kimi-Audio-7B-Instruct | 9.77B | 22.0 GB | 7.0 GB | 8GB GPU | Mac 16GB | Commercial OK | 2025-04-25 |
04Other families
Qwen · IBM Granite · NVIDIA Nemotron · Gemma · Mistral / Magistral / Devstral · GLM / CogVideoX (Z.ai) · DeepSeek · Ai2 OLMo / Molmo · Ant Ling / Ring / Ming · Liquid LFM · MiniCPM / VoxCPM · Tencent Hunyuan / Hy · Phi · Llama · Meituan LongCat · Jina embeddings & rerankers · Wan · Cohere Command / Aya · Xiaomi MiMo · FLUX · Microsoft VibeVoice / Fara / Harrier · InternLM / Intern-S · StepFun Step · MiniMax · Baidu ERNIE · NVIDIA Parakeet / Canary speech · Stable Diffusion / Stable Audio · Yi (01.AI) · ByteDance Seed · SmolLM / SmolVLM · BGE embeddings & rerankers · LTX Video · FunAudioLLM speech · gpt-oss · poolside Laguna · Whisper · Baichuan · Thinking Machines Inkling · xAI Grok · rednote dots
Totals include 10% + 0.5GB overhead and one 8K-token KV cache for text models. Estimates only. Method →