FreeLLMAll models › moonshot
Where to use it for free
- The lab's own chat: www.kimi.com (free with daily limits)
- Run it in one command with whichllm (free, open source):
uvx whichllm@latest plan "Kimi-K3.1-Pro"shows the GPU you need,uvx whichllm@latest run "Kimi-K3.1-Pro gguf"downloads it and starts a chat - Download and run it at home: official weights · GGUF files for Ollama and LM Studio
Arena scores
| Arena | Score | Rank | Votes |
|---|---|---|---|
| Agents | 5.3 agent score | #14 of 58 | 18K sessions |
Rank among all models, paid ones included. Source: Arena leaderboard dataset. The agent score is on a different scale (×100, from real agent sessions) and cannot be compared with the other scores.
Speed on popular hardware
| Hardware | GPU memory | Fits? | Speed |
|---|---|---|---|
| RTX 3060 12 GB | 12 GB | Does not fit | — |
| RTX 4060 Ti 16 GB | 16 GB | Does not fit | — |
| RTX 3090 | 24 GB | Does not fit | — |
| RTX 4090 | 24 GB | Does not fit | — |
| RTX 5090 | 32 GB | Does not fit | — |
| RX 7900 XTX | 24 GB | Does not fit | — |
| 2 × RTX 3090 | 48 GB | Does not fit | — |
| 4 × RTX 3090 | 96 GB | Does not fit | — |
| Mac M4 Max 128 GB | 96 GB | Does not fit | — |
| Ryzen AI Max+ 395 128 GB | 96 GB | Does not fit | — |
| 8 × H100 SXM | 640 GB | Does not fit | — |
Estimate at 4-bit with 32K context, 64 GB of DDR5 for offload. Speed ≈ memory bandwidth × 78% ÷ data read per token, plus a fixed cost per token, calibrated on 44 real llama.cpp tests; typical error about ±10%. Compare all GPUs.
GPU memory needed
| Level | 8K context | 32K context | 128K context | Quality vs full |
|---|---|---|---|---|
| Q3_K_M | 1392 GB | 1394 GB | 1405 GB | +4.7% noticeable loss |
| Q4_K_M | 1686 GB | 1689 GB | 1699 GB | +1.2% very small loss |
| Q5_K_M | 1978 GB | 1981 GB | 1991 GB | +0.4% near-lossless |
| Q6_K | 2281 GB | 2284 GB | 2294 GB | +0.2% near-lossless |
| Q8_0 | 2955 GB | 2958 GB | 2968 GB | ≈ 0% same as full |
| F16 | 5562 GB | 5564 GB | 5575 GB | ≈ 0% same as full |
Weights + context cache (from the model's architecture) + runtime. Quality: increase in perplexity against the full model, from llama.cpp tests, scaled by model size. Simulate your own setup.
Agents score over time
Arena score from 4 Jun 2026 to 4 Nov 2026.
Similar free models in Agents
- MiMo-V2.6-Pro 2.7 · 1024.2 B
- Gemini 3.8 Flash (High) 2.6
- DeepSeek-V4.1-Flash-20261017 (Max) 2.4
- GLM-5.3 (Max) 1.1 · 753.3 B
- Kimi K3 (Max) 0.8 · 2779.9 B
- Hy4 preview 0.7
Share this page:XLinkedInRedditWhatsApp
Spotted a wrong number or a broken link? Report a mistake