Up to date
moonshotai Kimi (Modified MIT) (modified-mit) out 10 months ago
1.03T parameters (8 of 384 experts per token), images and text in, text out, 256K tokens of context. 7 providers serve it from $0.45 per million input tokens.
Verified 2026-10-25 Rank #33 in LLMs · score 112.7 method v1 Sources 4 Downloads 30d 313.6k · #213 of 619 tracked
Picked by rule from 31 tracked pieces: the lab's own post, the most shared, a field report, the latest.
Kimi K2.5 self-directs an agent swarm with up to 100 sub-agents executing parallel workflows across up to 1,500 tool calls.
The measured 128-token workload contains 61,440 generated expert requests.
The 4 picks above, plus these 21. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.
Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.
5 providers with a live price, from $0.45 to $0.60 per million input tokens; the median is $0.53. One dot per provider, cheapest in teal.
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
Widely served (7 API providers); 39 quantized builds for local use; a big base for builders (75 fine-tunes, adapters and merges).
Top Spaces running it: akhaliq/anycoder ♥ 3,321 · hadadxyz/ai ♥ 819 · pliny-the-prompter/obliteratus ♥ 621 · hfmlsoc/different-flops ♥ 11
Weights only, an estimate: leave room for the KV cache. Method
About 616 GB at INT4, 1.23 TB at FP8, 2.46 TB at BF16: weights only, so leave room for the KV cache. Every expert must be resident, so total parameters count.
Estimate params × bytes per param × 1.2 · how we estimate
Third-party measurements only; author-reported numbers stay in Reference, marked as such.
70.8. Each tick is a model on this site with a measured SWE-bench Verified row; author-reported numbers stay in Reference. Percent resolved. agent: mini-SWE-agent, reasoning_effort=high
Source · leaderboard · 2026-02-17
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2026-01-01), amber triangles are 10 dated articles (each links to it), grey ticks are repo commits.
114 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Everything else, folded: the full tables, every source.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| measured SWE-bench Multilingual · Write code with agents | 67.3 percent resolved | agent: mini-SWE-agent | leaderboard | 2026-02-13 |
| measured SWE-bench Verified · Write code with agents | 70.8 percent resolved | agent: mini-SWE-agent, reasoning_effort=high | leaderboard | 2026-02-17 |
| author-reported AIME 2025 · Reason through hard problems | 96.1 | Thinking | model author | 2026-10-07 |
| author-reported CharXiv · Extract text from documents | 77.5 | Thinking | model author | 2026-10-07 |
| author-reported GPQA Diamond · Reason through hard problems | 87.6 | Thinking | model author | 2026-10-07 |
| author-reported HLE-Full | 30.1 | Thinking | model author | 2026-10-07 |
| author-reported HMMT · Reason through hard problems | 95.4 | Thinking | model author | 2026-10-07 |
| author-reported IMO-AnswerBench · Reason through hard problems | 81.8 | Thinking | model author | 2026-10-07 |
| author-reported InfoVQA · Extract text from documents | 92.6 | Thinking | model author | 2026-10-07 |
| author-reported MMLU-Pro · Reason through hard problems | 87.1 | Thinking | model author | 2026-10-07 |
| author-reported MMMU-Pro · Understand images and video | 78.5 | Thinking | model author | 2026-10-07 |
| author-reported MathVision · Understand images and video | 84.2 | Thinking | model author | 2026-10-07 |
| author-reported MathVista · Understand images and video | 90.1 | Thinking | model author | 2026-10-07 |
| author-reported OCRBench · Extract text from documents | 92.3 | Thinking | model author | 2026-10-07 |
| author-reported OmniDocBench · Extract text from documents | 88.8 | Thinking | model author | 2026-10-07 |
| author-reported SimpleVQA | 71.2 | Thinking | model author | 2026-10-07 |
| author-reported ZeroBench | 9 | Thinking | model author | 2026-10-07 |
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
Listed in LLMs
| Pipeline tag | image-text-to-text |
|---|---|
| Library | transformers |
| Architectures | KimiK25ForConditionalGeneration |
| Model type | kimi_k25 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 12,193,329,648 · F32: 23,040 · I32: 1,014,686,023,680 |
| Size band | XXL |
| MoE | yes |
| Experts | 384 |
| Experts per token | 8 |
| Context length | 262,144 (config.json max_position_embeddings) |
| Papers | arXiv:2602.02276 |
| License | Kimi (Modified MIT) (modified-mit) (other) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · KimiK25ForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · KimiK25ForConditionalGeneration · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Weights-only memory for Kimi-K2.5: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 2.46 TB · Cluster (≥640 GB) |
|---|---|
| FP8 (1 byte/param) | 1.23 TB · Cluster (≥640 GB) |
| INT4 (0.5 bytes/param) | 616 GB · Multi-GPU node (<640 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra. MoE: all experts must be resident, so total parameters count.
quantized: 39finetune: 45adapter: 29merge: 1
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| mlx-community/Kimi-K2.5 | quantized | — | 1.03T | XXL | other | mlx-community | 41 | 7M |
| nvidia/Kimi-K2.5-NVFP4 | quantized | NVFP4 | 518B | XXL | other | nvidia | 92 | 5.5M |
| amd/Kimi-K2.5-MXFP4 | quantized | MXFP4 | 550B | XXL | other | amd | 4 | 508.2k |
| lightseekorg/kimi-k2.5-eagle3 | finetune | — | 3.18B | M | mit | lightseekorg | 15 | 437.1k |
| unsloth/Kimi-K2.5-GGUF | quantized | GGUF | — | — | other | unsloth | 268 | 306.3k |
| lightseekorg/kimi-k2.5-eagle3-mla | finetune | — | 3.02B | M | other | lightseekorg | 1 | 120.9k |
| bartowski/moonshotai_Kimi-K2.5-GGUF | quantized | GGUF | — | — | other | bartowski | 12 | 46.2k |
| amd/Kimi-K2.5-MXFP4-AttnFP8 | quantized | MXFP4 | 550B | XXL | other | amd | 1 | 32.6k |
| nvidia/Kimi-K2.5-Thinking-Eagle3 | finetune | — | 1.84B | S | other | nvidia | 21 | 31k |
| QuantTrio/Kimi-K2.5-E304 | quantized | — | 841B | XXL | — | QuantTrio | 3 | 20.2k |
| nikita-savelyev-cerebras/tiny-random-kimi-k2.5 | finetune | — | 3.24M | S | — | nikita-savelyev-cerebras | 0 | 16.2k |
| QuantPasture/Kimi-K2.5-GGUF | quantized | GGUF | — | — | — | QuantPasture | 30 | 9.7k |
| spicyneuron/Kimi-K2.5-MLX-2.8bit | quantized | MLX | 1.03T | XXL | — | spicyneuron | 1 | 7.3k |
| mlx-community/Kimi-K2.5-mlx-DQ3_K_M-q8 | quantized | mlx | 1.03T | XXL | other | mlx-community | 3 | 6.5k |
| spicyneuron/Kimi-K2.5-MLX-2.5bit | quantized | MLX | 1.03T | XXL | — | spicyneuron | 1 | 6.3k |
| mlx-community/Kimi-K2.5-3bit | quantized | — | 1.03T | XXL | other | mlx-community | 0 | 6.1k |
| amd/Kimi-K2.5-W4A8 | quantized | — | 139B | XXL | other | amd | 2 | 5.8k |
| Drifter4242/Kimi-K2.5-vision | finetune | — | — | — | other | Drifter4242 | 0 | 5.7k |
| DevQuasar/moonshotai.Kimi-K2.5-GGUF | quantized | GGUF | — | — | — | DevQuasar | 2 | 5.3k |
| ubergarm/Kimi-K2.5-GGUF | quantized | GGUF | — | — | other | ubergarm | 11 | 4k |
| Ex0bit/Kimi-K2.5-PRISM-REAP-530B-A32B | quantized | — | 91.4B | XL | other | Ex0bit | 21 | 3.5k |
| unsloth/Kimi-K2.5 | quantized | — | 1.06T | XXL | other | unsloth | 14 | 3.5k |
| eatang/Kimi-K2.5-2layer | finetune | — | 20.4B | L | other | eatang | 0 | 2.9k |
| tiny-random/kimi-k2.5 | finetune | — | 3.24M | S | — | tiny-random | 4 | 2.2k |
| ubicloud/Kimi-K2.5-Pruned-20B | finetune | — | 20.9B | L | — | ubicloud | 1 | 2.3k |
| tbmod/Kimi-K2.5-MXFP4-mini | quantized | MXFP4 | 30.1B | L | other | tbmod | 0 | 1.9k |
| AxionML/Kimi-K2.5-MXFP8 | quantized | — | 1.03T | XXL | other | AxionML | 1 | 1.5k |
| eatang/Kimi-K2.5-2layer-BF16 | finetune | -BF16 | 20.4B | L | other | eatang | 0 | 1.1k |
| biosai-hf/kimi-k2.5-eagle3-mla | finetune | — | 3.02B | M | other | biosai-hf | 0 | 1k |
| bakosh/Huihui-Kimi-K2.5-BF16-abliterated-GGUF | quantized | -BF16 | — | — | other | bakosh | 1 | 977 |
| yujiepan/kimi-k2.5-tiny-random | finetune | — | 3.24M | S | — | yujiepan | 1 | 893 |
| lhca521/Huihui-Kimi-K2.5-BF16-abliterated-GGUF | quantized | -BF16 | — | — | other | lhca521 | 1 | 849 |
| DripGhad1223/Huihui-Kimi-K2.5-BF16-abliterated-GGUF | quantized | -BF16 | — | — | other | DripGhad1223 | 1 | 736 |
| ISTA-DASLab/Kimi-K2.5-P48NVFP4-MoESQ | quantized | — | 646B | XXL | other | ISTA-DASLab | 1 | 628 |
| rbinrs/rbinrs-Kimi-K2.5-BF16-abliterated-GGUF | quantized | -BF16 | — | — | other | rbinrs | 0 | 668 |
| syntheticlab/Kimi-K2.5-NVFP4A16 | quantized | — | 166B | XXL | other | syntheticlab | 1 | 610 |
| promotewillgo/apollo5GWKiiudZ5AePoLA6HxH9HwF3p8Q98o1Q6tRKepsp4EouRak | quantized | — | 1.03T | XXL | other | promotewillgo | 0 | 492 |
| kepom/rbinrs-Kimi-K2.5-BF16-abliterated-GGUF | quantized | -BF16 | — | — | other | kepom | 0 | 326 |
| Panchovix/Kimi-K2.5-GGUF | quantized | GGUF | — | — | — | Panchovix | 0 | 323 |
| Cirrascale/Kimi-K2.5-NVFP4 | quantized | NVFP4 | — | — | other | Cirrascale | 0 | 301 |
| Date | Commit | SHA |
|---|---|---|
| 2026-04-30 | support-transformers-inference (#115) | 4d01dfe |
| 2026-02-27 | fix recommended order of multimodal inputs | 54383e8 |
| 2026-02-05 | doc: add link for paper and github repo (#53) | 3367c8d |
| 2026-02-03 | Update README.md | c0d6821 |
| 2026-02-03 | link to hf paper | df44e69 |
| 2026-01-31 | Updated deployment command to include encoder TP mode | 2426b45 |
| 2026-01-30 | Remove unnecessary deprecated import (#26) | 61d22d6 |
| 2026-01-29 | Remove default system prompt, fix media_start token and update readme (#20) | 6650964 |
| 2026-01-29 | docs: fix vLLM tensor parallel flag in deploy guide (#24) | 907b3ee |
| 2026-01-28 | Fix grammar and punctuation errors in README.md (#12) | 8d93622 |
| 2026-01-28 | Add compressed-tensors tag (#9) | 911dce1 |
| 2026-01-28 | Compatibility with transformers v5 (#11) | 8922bd6 |
| 2026-01-27 | Update README.md | 9659d5d |
| 2026-01-27 | Update README.md | a13437f |
| 2026-01-27 | Update README.md | 9196113 |
| 2026-01-27 | Update README.md | 1336cf3 |
| 2026-01-27 | Update README.md | a7eaad5 |
| 2026-01-27 | Update README.md | 1a2ecac |
| 2026-01-27 | Update README.md | 7eef118 |
| 2026-01-01 | Add files using upload-large-folder tool | 70a8ad0 |
| 2026-01-01 | initial commit | 511c492 |
Hosted endpoints observed for Kimi-K2.5, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting. Cheapest observed input price: $0.45/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| DeepInfra hf | moonshotai/Kimi-K2.5 | 256K | — | $0.45 | $2.25 | 86.5 | 494 ms | yes | live | 2026-10-25 |
| AtlasCloud openrouter | moonshotai/kimi-k2.5 | 256K | int4 | $0.49 | $2.50 | — | — | yes | status -2 | 2026-10-24 |
| SiliconFlow openrouter | moonshotai/kimi-k2.5 | 256K | int4 | $0.50 | $2.50 | — | — | yes | live | 2026-10-24 |
| Venice openrouter | moonshotai/kimi-k2.5 | 250K | unknown | $0.53 | $3.33 | — | — | yes | live | 2026-10-24 |
| Novita hf, openrouter | moonshotai/kimi-k2.5 | 256K | unknown | $0.57 | $2.85 | 95 | 1395 ms | yes | live | 2026-10-25 |
| Amazon Bedrock openrouter | moonshotai/kimi-k2.5 | 256K | unknown | $0.60 | $3.00 | — | — | yes | live | 2026-10-24 |
| Featherless hf | moonshotai/Kimi-K2.5 | — | — | — | — | — | — | — | live | 2026-10-25 |
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:32 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 14:59 UTC · Last verified 2026-10-25
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.
