- Size
- 1.0T parametersfrom the GGUF metadata of a conversion, because this repo's own weights are packed: unsloth/Kimi-K2-Thinking-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
- 4 Nov 2025the date the repo was created
- Context length
- 256k tokensfrom the model's config
- Kind
- Base model (no base model declared on the Hub)DeepseekV3ForCausalLM
- On Hugging Face
- moonshotai/Kimi-K2-Thinking59k downloads in the last 30 days, 1.7k likes
Scores
- default setting21.4%±2.45 Dec 2025, log
Epoch's name: moonshotai/Kimi-K2-Thinking. Best closed model on this set: gpt-5.5-pro-pre-release_high, 52.4% (OpenAI). All 101 results
- default setting0.0%4 Dec 2025, log
Epoch's name: moonshotai/Kimi-K2-Thinking. Best closed model on this set: gdm-ai-co-mathematician, 47.9% (Google DeepMind). All 72 results
- default setting17.6%via Epoch AI
- default setting11.9%via Epoch AI
Epoch's name: moonshotai/Kimi-K2-Thinking, kimi-k2-thinking. Best closed model on this set: gpt-5.4-2026-03-05_xhigh, 27.9% (OpenAI). All 23 results
- default setting59.2%METR - Measuring AI Ability to Complete Long Tasks
Epoch's name: kimi-k2-thinking. Best closed model on this set: claude-mythos-preview-early, 85.2% (Anthropic). All 49 results
- default setting39.6%simple-bench.com
Epoch's name: kimi-k2-thinking. Best closed model on this set: claude-fable-5-2_unknown, 94.9% (Anthropic). All 125 results
- Agent: Terminus 235.7%±2.86 Nov 2025, tbench.ai
- Agent: Terminus 235.7%±2.811 Nov 2025, Terminal-Bench v2 Leaderboard
Epoch's name: kimi-k2-thinking. Best closed model on this set: gpt-5.5_unknown, 84.7% (OpenAI). All 202 results
- default setting42.8%±0.0htihle.github.io
Epoch's name: kimi-k2-thinking. 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-Thinking), 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-Thinking-GGUF (converted by unsloth), 3.4k downloads. Sizes are the real files on the Hub, all parts added up.
| Format | Size | Memory needed | Get it |
|---|---|---|---|
| UD-IQ3_XXS | 392.7 GBfile size, 9 parts | 393.8 GB392.7 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:UD-IQ3_XXSDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "UD-IQ3_XXS/Kimi-K2-Thinking-UD-IQ3_XXS-*-of-00009.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| Q3_K_S | 412.7 GBfile size, 10 parts | 413.8 GB412.7 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:Q3_K_SDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "Q3_K_S/Kimi-K2-Thinking-Q3_K_S-*-of-00010.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| UD-Q3_K_XL | 423.9 GBfile size, 10 parts | 424.9 GB423.9 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:UD-Q3_K_XLDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "UD-Q3_K_XL/Kimi-K2-Thinking-UD-Q3_K_XL-*-of-00010.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| Q3_K_M | 456.3 GBfile size, 11 parts | 457.4 GB456.3 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:Q3_K_MDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "Q3_K_M/Kimi-K2-Thinking-Q3_K_M-*-of-00011.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| IQ4_XS | 509.6 GBfile size, 12 parts | 510.6 GB509.6 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:IQ4_XSDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "IQ4_XS/Kimi-K2-Thinking-IQ4_XS-*-of-00012.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| IQ4_NL | 539.3 GBfile size, 12 parts | 540.3 GB539.3 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:IQ4_NLDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "IQ4_NL/Kimi-K2-Thinking-IQ4_NL-*-of-00012.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| Q4_0 | 541.3 GBfile size, 13 parts | 542.3 GB541.3 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:Q4_0Download hf download unsloth/Kimi-K2-Thinking-GGUF --include "Q4_0/Kimi-K2-Thinking-Q4_0-*-of-00013.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| Q4_K_S | 543.2 GBfile size, 13 parts | 544.3 GB543.2 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:Q4_K_SDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "Q4_K_S/Kimi-K2-Thinking-Q4_K_S-*-of-00013.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| Q4_K_M | 578.6 GBfile size, 13 parts | 579.6 GB578.6 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:Q4_K_MDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "Q4_K_M/Kimi-K2-Thinking-Q4_K_M-*-of-00013.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| Q4_1 | 598.8 GBfile size, 13 parts | 599.8 GB598.8 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:Q4_1Download hf download unsloth/Kimi-K2-Thinking-GGUF --include "Q4_1/Kimi-K2-Thinking-Q4_1-*-of-00013.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| UD-Q4_K_XL | 601.9 GBfile size, 14 parts | 602.9 GB601.9 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:UD-Q4_K_XLDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "UD-Q4_K_XL/Kimi-K2-Thinking-UD-Q4_K_XL-*-of-00014.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| Q5_K_S | 658.5 GBfile size, 15 parts | 659.5 GB658.5 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:Q5_K_SDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "Q5_K_S/Kimi-K2-Thinking-Q5_K_S-*-of-00015.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| Q5_K_M | 678.7 GBfile size, 16 parts | 679.7 GB678.7 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:Q5_K_MDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "Q5_K_M/Kimi-K2-Thinking-Q5_K_M-*-of-00016.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| UD-Q5_K_XL | 679.7 GBfile size, 16 parts | 680.7 GB679.7 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:UD-Q5_K_XLDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "UD-Q5_K_XL/Kimi-K2-Thinking-UD-Q5_K_XL-*-of-00016.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| Q6_K | 785.0 GBfile size, 18 parts | 786.1 GB785.0 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:Q6_KDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "Q6_K/Kimi-K2-Thinking-Q6_K-*-of-00018.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| UD-Q6_K_XL | 818.7 GBfile size, 19 parts | 819.7 GB818.7 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:UD-Q6_K_XLDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "UD-Q6_K_XL/Kimi-K2-Thinking-UD-Q6_K_XL-*-of-00019.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| Q8_0 | 1016.1 GBfile size, 23 parts | 1017.2 GB1016.1 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:Q8_0Download hf download unsloth/Kimi-K2-Thinking-GGUF --include "Q8_0/Kimi-K2-Thinking-Q8_0-*-of-00023.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| UD-Q8_K_XL | 1108.3 GBfile size, 25 parts | 1109.4 GB1108.3 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:UD-Q8_K_XLDownload hf download unsloth/Kimi-K2-Thinking-GGUF --include "UD-Q8_K_XL/Kimi-K2-Thinking-UD-Q8_K_XL-*-of-00025.gguf" --local-dir Kimi-K2-Thinking-GGUF |
| BF16 | 1912.2 GBfile size, 46 parts | 1913.2 GB1912.2 + 0.54 + 0.50 | Commandsllama.cpp llama-server -hf unsloth/Kimi-K2-Thinking-GGUF:BF16Download hf download unsloth/Kimi-K2-Thinking-GGUF --include "BF16/Kimi-K2-Thinking-BF16-*-of-00046.gguf" --local-dir Kimi-K2-Thinking-GGUF |
MLX for Apple silicon
| Format | Size | Memory needed | Repo |
|---|---|---|---|
| MLXmlx-community/Kimi-K2-Thinking | 612.5 GBfile size | 613.5 GB612.5 + 0.54 + 0.50 | Download page |
| MLX 4bitmlx-community/Kimi-K2-Thinking-4bit | 1150.4 GBfile size | 1151.4 GB1150.4 + 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-Thinking --local-dir Kimi-K2-ThinkingOther conversions (6; sizes not read)
- nvidia/Kimi-K2-Thinking-NVFP4 NVFP4, 265k downloads
- unsloth/Kimi-K2-Thinking Other, 174 downloads
- amd/Kimi-K2-Thinking-MXFP4 Other, mxfp4, 128 downloads
- amd/Kimi-K2-Thinking-W4A8 Other, 127 downloads
- amd/Kimi-K2-Thinking-MXFP4-AttnFP8 Other, mxfp4, 125 downloads
- luisra/Kimi-K2-Thinking-4bit bitsandbytes, 4bit, 45 downloads
The Hub tags 18 quantised repos as conversions of this model; we read the most downloaded and show the maker's own first.
Run it by API
2 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 |
|---|---|---|---|---|---|---|---|---|---|
| Novita | $0.60 | $2.50 | 256k | bf16 | 37 | 1.1 s | 100.0% | 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-Thinking:novita", "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-Thinking-Eagle31.8B280Unreviewed
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.