Up to date
moonshotai Kimi (Modified MIT) (modified-mit) out 12 months ago
1.03T parameters (8 of 384 experts per token), text in, text out, 256K tokens of context. 3 providers serve it from $0.60 per million input tokens.
Verified 2026-10-25 Rank #129 in LLMs · score 94.7 method v1 Sources 4 Downloads 30d 47.7k · #386 of 619 tracked
Picked by rule from 23 tracked pieces: the lab's own post, the most shared, a field report, the latest.
Kimi K2 Thinking is an open-source thinking agent model that executes up to 300 sequential tool calls to solve complex problems.
Kimi K2.7 Code is generally available in GitHub Copilot as the first open-weight model offered as a selectable option in the Copilot model picker.
The 3 picks above, plus these 20. 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.
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.
Served by 3 API providers; 18 quantized builds for local use.
Top Spaces running it: akhaliq/anycoder ♥ 3,321 · hadadxyz/ai ♥ 819 · umint/ai ♥ 258 · llamameta/Free-kimi-k2-thinking ♥ 16
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.
63.4. 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
Source · leaderboard · 2025-12-10
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2025-11-04), amber triangles are 8 dated articles (each links to it), grey ticks are repo commits.
41 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 |
|---|---|---|---|---|
| author-reported SciCode | 44.8 | no tools | model author | 2026-10-07 |
| author-reported Seal-0 | 56.3 | w/ tools | model author | 2026-10-07 |
| author-reported Terminal-Bench · Write code with agents | 47.1 | w/ simulated tools (JSON) | 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.
| Pipeline tag | text-generation |
|---|---|
| Library | transformers |
| Architectures | DeepseekV3ForCausalLM |
| Model type | kimi_k2 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 11,722,193,536 · 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) |
| License | Kimi (Modified MIT) (modified-mit) (other) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · DeepseekV3ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · DeepseekV3ForCausalLM · 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-Thinking: 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: 18finetune: 12adapter: 10merge: 1
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| nvidia/Kimi-K2-Thinking-NVFP4 | quantized | NVFP4 | 519B | XXL | other | nvidia | 40 | 652k |
| mlx-community/Kimi-K2-Thinking | quantized | — | 1.03T | XXL | other | mlx-community | 17 | 122.2k |
| unsloth/Kimi-K2-Thinking-GGUF | quantized | GGUF | — | — | other | unsloth | 115 | 88.9k |
| ubergarm/Kimi-K2-Thinking-GGUF | quantized | GGUF | — | — | other | ubergarm | 22 | 19.1k |
| bartowski/moonshotai_Kimi-K2-Thinking-GGUF | quantized | GGUF | — | — | other | bartowski | 5 | 14.7k |
| mlx-community/Kimi-K2-Thinking-4bit | quantized | -4bit | 1.03T | XXL | other | mlx-community | 9 | 11.1k |
| DevQuasar/moonshotai.Kimi-K2-Thinking-GGUF | quantized | GGUF | — | — | — | DevQuasar | 5 | 7.9k |
| amd/Kimi-K2-Thinking-MXFP4 | quantized | MXFP4 | 550B | XXL | other | amd | 0 | 7.8k |
| moxin-org/Kimi-K2-Thinking-Moxin-GGUF | quantized | GGUF | — | — | other | moxin-org | 9 | 7.3k |
| amd/Kimi-K2-Thinking-MXFP4-AttnFP8 | quantized | MXFP4 | 550B | XXL | other | amd | 0 | 3.3k |
| mlx-community/Kimi-K2-Thinking-3bit | quantized | — | 1.03T | XXL | other | mlx-community | 1 | 2.4k |
| unsloth/Kimi-K2-Thinking-BF16 | finetune | -BF16 | — | — | other | unsloth | 6 | 1.8k |
| squ11z1/Hypnos-Colossus-1T | quantized | GGUF | — | — | other | squ11z1 | 2 | 1.9k |
| amd/Kimi-K2-Thinking-W4A8 | quantized | — | — | — | other | amd | 0 | 1.9k |
| unsloth/Kimi-K2-Thinking | quantized | — | 1.06T | XXL | other | unsloth | 4 | 1.5k |
| picklass/Kimi-K2-Thinking-MLX-4.25bit | quantized | MLX | — | — | — | picklass | 0 | 1.5k |
| nvidia/Kimi-K2-Thinking-Eagle3 | finetune | — | 1.84B | S | other | nvidia | 2 | 995 |
| DevQuasar-2/moonshotai.Kimi-K2-Thinking-BF16 | finetune | -BF16 | 1.03T | XXL | — | DevQuasar-2 | 0 | 851 |
| ling1000T/Kimi-K2-Thinking-gguf | quantized | gguf | — | — | other | ling1000T | 1 | 500 |
| sdan/jokegen2-1t-rl | adapter | — | — | — | apache-2.0 | sdan | 9 | 0 |
| mradermacher/Kimi-K2-Thinking-i1-GGUF | quantized | GGUF | — | — | other | mradermacher | 3 | 295 |
| luisra/Kimi-K2-Thinking-4bit | quantized | -4bit | — | — | other | luisra | 0 | 202 |
| axs2k/anubis | finetune | — | — | — | gpl-3.0 | axs2k | 3 | 0 |
| sdan/jokegen2-1t-sft | adapter | — | — | — | apache-2.0 | sdan | 3 | 0 |
| UNYIL17/MODEL-ONE-1 | merge | — | — | — | apache-2.0 | UNYIL17 | 1 | 0 |
| obalcells/kimi-k2-thinking-mo-midtrained-lora | adapter | — | — | — | — | obalcells | 0 | 50 |
| uwuwuwuwuwuwu/kimi-k2.5-reward-hacking-step-0 | adapter | — | — | — | — | uwuwuwuwuwuwu | 0 | 49 |
| d0ks/perry-anderson-lora | adapter | — | — | — | apache-2.0 | d0ks | 0 | 39 |
| mradermacher/Kimi-K2-Thinking-GGUF | finetune | GGUF | — | — | other | mradermacher | 0 | 0 |
| Mashbox41/faith2 | finetune | — | — | — | apache-2.0 | Mashbox41 | 0 | 0 |
| guitar73/guitar | adapter | — | — | — | openrail | guitar73 | 0 | 0 |
| firdavsCreative2025/FirdavsDijital-text-classifier | finetune | — | — | — | mit | firdavsCreative2025 | 0 | 0 |
| samgiz/docc_e | finetune | — | — | — | apache-2.0 | samgiz | 0 | 0 |
| ADOBET88/pildun2026 | finetune | — | — | — | apache-2.0 | ADOBET88 | 0 | 0 |
| wajdi-26/hoom | finetune | — | — | — | apache-2.0 | wajdi-26 | 0 | 0 |
| BACCHUS45/Potex | adapter | — | — | — | apache-2.0 | BACCHUS45 | 0 | 0 |
| Maxmaxjawa/Scripshit | finetune | — | — | — | llama3.3 | Maxmaxjawa | 0 | 0 |
| arvindcr4/tinker-rl-bench-arch_gsm8k_kimi-k2 | finetune | — | — | — | mit | arvindcr4 | 0 | 0 |
| arvindcr4/tinker-rl-arch_gsm8k_kimi-k2-kimi-k2 | adapter | — | — | — | apache-2.0 | arvindcr4 | 0 | 0 |
| aidenkbm/jokegen2-1t-rl | adapter | — | — | — | apache-2.0 | aidenkbm | 0 | 0 |
| Date | Commit | SHA |
|---|---|---|
| 2026-01-30 | Transformers v5 support (#53) | a51ccc0 |
| 2025-11-08 | fix-default-system-prompt (#12) | 6126819 |
| 2025-11-07 | Fix args in SGLang launch command (#9) | f1ed9ca |
| 2025-11-07 | Update README.md | 1b9dbb7 |
| 2025-11-07 | fix vocab size (#4) | 6e3cdad |
| 2025-11-06 | Update LICENSE | 357b94a |
| 2025-11-06 | Update README.md | 603577f |
| 2025-11-06 | Update README.md | 7d2962b |
| 2025-11-06 | Update README.md | f5ed4a8 |
| 2025-11-06 | Update README.md | d02605f |
| 2025-11-06 | Update README.md | 5b64d2a |
| 2025-11-06 | update | c9caddd |
| 2025-11-06 | update | f1cdb45 |
| 2025-11-06 | update deploy guidance | 6dead1a |
| 2025-11-06 | fix tokenizer | a060a3e |
| 2025-11-06 | Update README.md | 22bbb72 |
| 2025-11-06 | Update README.md | 7b23274 |
| 2025-11-04 | initial commit | 652f5f9 |
Hosted endpoints observed for Kimi-K2-Thinking, 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: $0.60/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| Google openrouter | moonshotai/kimi-k2-thinking | 256K | unknown | $0.60 | $2.50 | — | — | yes | status -5 | 2026-10-24 |
| Novita openrouter | moonshotai/kimi-k2-thinking | 256K | bf16 | $0.60 | $2.50 | — | — | yes | live | 2026-10-24 |
| Featherless hf | moonshotai/Kimi-K2-Thinking | — | — | — | — | — | — | — | 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.
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:35 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:02 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.
