- Size
- 1.0T parametersfrom the GGUF metadata of a conversion, because this repo's own weights are packed: unsloth/Kimi-K2.6-GGUF
- Licence
- unknown, MiniMax Model LicenseOur reading, check the licence: Despite the id, this text is the MiniMax Model License (dated for MiniMax-M2.5), not a modified MIT licence: it lets you use, copy, change and distribute the model with no limit on commercial use and no user or revenue threshold. If you distribute the model or a product or service that uses it, you must include the licence, mark changed files and keep a Notice file, though you may license your own changes differently. A Prohibited Uses Policy is part of the licence and bars, among other things, any military purpose and fully automated decisions that adversely affect someone's legal rights.
- On the Hub since
- 14 Apr 2026the date the repo was created
- Context length
- 256k tokensfrom the model's config
- Kind
- Base model (no base model declared on the Hub)KimiK25ForConditionalGeneration
- On Hugging Face
- moonshotai/Kimi-K2.6199k downloads in the last 30 days, 1.6k likes
Scores
- default setting20.0%±4.013 Oct 2026, log
Epoch's name: kimi-k2.6. Best closed model on this set: claude-fable-5-2_max, 79.0% (Anthropic). All 166 results
- default setting2.4%25 Jun 2026, Epoch AI
Epoch's name: kimi-k2.6. Best closed model on this set: gpt-6.2-astra_max, 92.4% (OpenAI). All 33 results
- default setting39.0%±2.97 May 2026, Epoch AI
Epoch's name: kimi-k2.6. Best closed model on this set: gpt-5.5-pro-pre-release_high, 52.4% (OpenAI). All 101 results
- default setting14.6%±5.18 May 2026, Epoch AI
Epoch's name: kimi-k2.6. Best closed model on this set: gdm-ai-co-mathematician, 47.9% (Google DeepMind). All 72 results
- default setting25.6%±7.110 Jun 2026, log
Epoch's name: kimi-k2.6. Best closed model on this set: claude-fable-5-2_max, 100.0% (Anthropic). All 83 results
- default setting57.2%±2.910 Jun 2026, log
Epoch's name: kimi-k2.6. Best closed model on this set: claude-fable-5-2_max, 95.1% (Anthropic). All 127 results
- default setting21.7%±5.211 Sep 2026, Epoch AI
Epoch's name: kimi-k2.6. Best closed model on this set: claude-fable-5-2_max, 90.0% (Anthropic). All 41 results
- default setting90.8%±1.71 May 2026, log
Epoch's name: kimi-k2.6. Best closed model on this set: gpt-6.2-astra_max, 96.8% (OpenAI). All 341 results
Epoch's name: kimi-k2.6. Best closed model on this set: gpt-6.2-astra_max, 95.0% (OpenAI). All 148 results
- default setting96.1%±2.42 May 2026, log
Epoch's name: kimi-k2.6. Best closed model on this set: claude-fable-5_high, 100.0% (Anthropic). All 318 results
- default setting34.9%±1.510 Aug 2026, Epoch AI
Epoch's name: kimi-k2.6. Best closed model on this set: claude-fable-5-2_max, 94.0% (Anthropic). All 105 results
- default setting76.7%±1.98 May 2026, log
Epoch's name: kimi-k2.6. Best closed model on this set: claude-opus-4-7_max, 83.5% (Anthropic). All 35 results
- default setting3.9%andonlabs.com
Epoch's name: kimi-k2.6. Best closed model on this set: gemini-4-carbon_unknown, 58.6% (Google DeepMind). All 38 results
- default setting8.0%via Epoch AI
Epoch's name: kimi-k2.6. Best closed model on this set: gemini-4-carbon_high, 36.0% (Google DeepMind). All 271 results
- default setting90.9%conceptualreasoning.ai
Epoch's name: kimi-k2.6. Best closed model on this set: claude-fable-5-2_unknown, 99.2% (Anthropic). All 241 results
- default setting18.4%via Epoch AI
- default setting16.3%via Epoch AI
Epoch's name: kimi-k2.6. Best closed model on this set: claude-mythos-preview-early, 73.8% (Anthropic). All 20 results
- default setting0.9%1 May 2026, gbaeval.com
Epoch's name: kimi-k2.6. Best closed model on this set: claude-opus-5-5_unknown, 90.6% (Anthropic). All 28 results
- default setting37.3%32.8 to 41.7conceptualreasoning.ai
Epoch's name: kimi-k2.6. Best closed model on this set: claude-fable-5-2_xhigh, 70.6% (Anthropic). All 203 results
- Reasoning: enabled; Tool setting: standard; Step budget: 5004.6%osworld-v2.xlang.ai
Epoch's name: kimi-k2.6. Best closed model on this set: claude-opus-5_max, 31.4% (Anthropic). All 16 results
- default setting16.0%±3.7vals.ai
Epoch's name: kimi-k2.6. Best closed model on this set: claude-fable-5-1_max, 100.0% (Anthropic). All 85 results
- default setting55.9%±0.0htihle.github.io
Epoch's name: kimi-k2.6. Best closed model on this set: gpt-6-astra_promax, 93.6% (OpenAI). All 173 results
Claimed by the maker Claimed
The model card has no claimed scores in a machine-readable form (the card's model-index), so none are shown. The maker may give scores in the card text: read the card.
Other views, which we link to and do not copy: look this model up on Artificial Analysis and the Arena leaderboard.
Independent scores from Epoch AI (CC-BY 4.0), fetched 19 Nov 2026: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from ‘https://epoch.ai/benchmarks’ [online resource]. Each row links its own source. What the badges mean.
Run it yourself
Memory needed = weights + context cache + 0.50 GB for the runtime. Context cache at 8k tokens: 0.54 GB, worked out from the model's config (moonshotai/Kimi-K2.6), at 16-bit precision. Choose a machine above to see a verdict for each format. Method
GGUF for llama.cpp, Ollama, LM Studio and others
unsloth/Kimi-K2.6-GGUF (converted by unsloth), 4.2k downloads. Sizes are the real files on the Hub, all parts added up. For image input you also need the projector file mmproj-F16.gguf (0.89 GB), not counted below.
| Format | Size | Memory needed | Get it |
|---|---|---|---|
| UD-Q2_K_XL | 317.1 GBfile size, 8 parts | 318.1 GB317.1 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2.6-GGUF:UD-Q2_K_XLDownload hf download unsloth/Kimi-K2.6-GGUF --include "UD-Q2_K_XL/Kimi-K2.6-UD-Q2_K_XL-*-of-00008.gguf" --local-dir Kimi-K2.6-GGUF |
| UD-Q4_K_XL | 543.6 GBfile size, 14 parts | 544.7 GB543.6 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2.6-GGUF:UD-Q4_K_XLDownload hf download unsloth/Kimi-K2.6-GGUF --include "UD-Q4_K_XL/Kimi-K2.6-UD-Q4_K_XL-*-of-00014.gguf" --local-dir Kimi-K2.6-GGUF |
| UD-Q8_K_XL | 553.7 GBfile size, 14 parts | 554.7 GB553.7 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2.6-GGUF:UD-Q8_K_XLDownload hf download unsloth/Kimi-K2.6-GGUF --include "UD-Q8_K_XL/Kimi-K2.6-UD-Q8_K_XL-*-of-00014.gguf" --local-dir Kimi-K2.6-GGUF |
| BF16 | 1912.2 GBfile size, 46 parts | 1913.2 GB1912.2 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2.6-GGUF:BF16Download hf download unsloth/Kimi-K2.6-GGUF --include "BF16/Kimi-K2.6-BF16-*-of-00046.gguf" --local-dir Kimi-K2.6-GGUF |
MLX for Apple silicon
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| MLX 6bitspicyneuron/Kimi-K2.6-MLX-3.6bit | 427.6 GBfile size | 428.6 GB427.6 + 0.54 + 0.50 | Download page |
| MLXmlx-community/Kimi-K2.6-mlx-DQ3_K_M-q8 | 438.3 GBfile size | 439.3 GB438.3 + 0.54 + 0.50 | Download page |
Original weights
The maker's own weights are packed into integer tensors, so the safetensors header gives no size we can trust; the repo lists the real file sizes.
hf download moonshotai/Kimi-K2.6 --local-dir Kimi-K2.6Other conversions (10; sizes not read)
- nvidia/Kimi-K2.6-NVFP4 NVFP4, 148k downloads
- amd/Kimi-K2.6-MXFP4 Other, mxfp4, 6.6k downloads
- RedHatAI/Kimi-K2.6-FP8-BLOCK FP8, 940 downloads
- RedHatAI/Kimi-K2.6-NVFP4 NVFP4, 303 downloads
- unsloth/Kimi-K2.6 Other, 178 downloads
- gratex/Kimi-K2.6-REAM-25 Other, 166 downloads
- amd/Kimi-K2.6-NVFP4 NVFP4, 118 downloads
- intel-ai/Kimi-K2.6-MXFP4-FP8KV-CT-AutoRound Other, mxfp4, 75 downloads
- festr2/kimi-k2.6-eagle3-mla-fp8 FP8, 67 downloads
- niraj-runara/Kimi-K2.6-NVFP4-fused NVFP4, 67 downloads
The Hub tags 45 quantised repos as conversions of this model; we read the most downloaded and show the maker's own first.
Run it by API
12 providers serve this model through the Hugging Face router. Prices are US dollars per million tokens. Precision is what the provider reports; a lower precision than the original weights can change answers. Live first, then cheapest output.
| Provider | In | Out | Context | Precision | Tokens/s | First token | Uptime 24h | Tools | JSON |
|---|---|---|---|---|---|---|---|---|---|
| DigitalOcean | $0.57 | $2.40 | 256k | not given | 48 | 0.9 s | 100.0% | yes | yes |
| Inceptron | $0.58 | $2.87 | 256k | int4 | 78 | 0.5 s | 100.0% | yes | yes |
| Novita | $0.80 | $3.40 | 256k | not given | 31 | 4.0 s | 97.4% | yes | yes |
| SiliconFlow | $0.77 | $3.40 | 256k | fp8 | 21 | 0.9 s | 100.0% | yes | yes |
| CoreWeave | $0.65 | $3.41 | 256k | fp4 | 113 | 0.8 s | 99.8% | yes | yes |
| Crusoe | $0.70 | $3.50 | 256k | bf16 | 70 | 0.4 s | 99.3% | yes | yes |
| DeepInfra | $0.75 | $3.50 | 256k | fp4 | 20 | 0.8 s | 98.6% | yes | yes |
| Parasail | $0.75 | $3.50 | 256k | int4 | 75 | 0.6 s | 99.9% | yes | yes |
| Venice | $0.75 | $3.50 | 250k | int4 | 24 | 0.7 s | 89.4% | yes | yes |
| Cloudflare | $0.95 | $4.00 | 256k | not given | 53 | 0.9 s | 100.0% | yes | yes |
| Moonshot AIthe maker | $0.95 | $4.00 | 256k | not given | 30 | 1.3 s | 99.5% | yes | yes |
| featherless-ai | ? | ? | ? | not given | ? | ? | ? | not given | not given |
Through the router, with a Hugging Face token (the router bills your Hugging Face account). Change the name after the colon to pick another provider.
curl https://router.huggingface.co/v1/chat/completions \
-H "Authorization: Bearer $HF_TOKEN" -H "Content-Type: application/json" \
-d '{"model": "moonshotai/Kimi-K2.6:digitalocean", "messages": [{"role": "user", "content": "Hello"}]}'Not built yet: one key for every open model, with BlockRun, and our own tested-quality score per provider. Until then, precision and measured speed are the best guide to a provider here.
Fine-tunes of this model
Repos that declare this model as their base, most downloaded first. Listing is not a recommendation; unreviewed ones have not been checked.
- nvidia/Kimi-K2.6-Eagle31.8B755Unreviewed
- nvidia/Kimi-K2.6-DFlash3.5B383Unreviewed
Planned for this page Not built yet
- Try it in private chat
- A link to Solana AI's confidential chat for the models it serves. This model is not listed as served there.
- Fine-tune it
- Fine-tuning a few base models from this page and publishing the result here.
- Creators and earnings
- Pages for people who publish fine-tunes and adapters of this model, and what they earn.
- Fund and bounties
- Bounties for new open models, paid from a public fund. How the fund will work.
- Contribute compute
- How to give compute to open model work, with Akash.
Model data from the Hugging Face Hub 19 Nov 2026, 13:43 UTC. The repo. Found a mistake on this page? Tell us.