Support · Kimi K2
Will it last?
Ownership levelPartialnone·limited·partial·substantial·fullAnalytical input C ยท 62.8/100
This page is a projection of the one entry record, the Transparency factor that Support covers. The full verdict is set by all four factors together, floor-weighted so the weakest caps the whole.
Which domain expands which factor
- AssessUse & modify + Transparency
- ImplementData control + Reliability
- UseReliability
- SupportTransparency
Common problems & fixes
Common pitfalls, not an exhaustive catalogue:
- Out-of-memory and sharding failures: the 1T model must be sharded across a multi-GPU or
multi-node cluster. Size tensor and pipeline parallelism to the hardware, or use KTransformers
CPU offload.
- Block-FP8 kernel or engine errors: match the vLLM, SGLang or TensorRT-LLM version to one that
supports the block-FP8 weights and MLA attention. Version mismatch is a common cause.
- Garbled or over-verbose Instruct output: the chat template is most likely not applied. Use
apply_chat_template and do not prompt the Base checkpoint as a chat model.
- Non-reproducible results or silent updates: a deployment floating on
main changes under you.
Pin an exact revision and verify checksums.
Versions, changelog & cadence
Kimi K2 ships as Base and Instruct checkpoints on the verified moonshotai org, tracked as
immutable Hugging Face revisions. The revision hash is the changelog anchor: pin it and diff
against a newer revision when upgrading. The technical report "Kimi K2: Open Agentic
Intelligence" accompanies the release. The Instruct card marks Kimi-K2-Instruct-0905 as the
newer version. Later generations (K2.5, K2.6, K2.7-Code and K3) exist and are not covered here.
On the hosted API, K2 series models closed to new users on 2026-10-31 and are planned for
retirement by the end of December 2026; the downloadable weights are not affected.
Usage questions go to the model discussion tabs on the moonshotai Hugging Face org
(ev-kimi-hf). Deployment, serving-engine and reproduction problems go to GitHub issues on
moonshotai/Kimi-K2 (ev-github). There is no paid support tier for the open weights; support
is community and maintainer based.
Tracked known issues
These come from model-card and third-party caveats, not a formal issue tracker, so coverage is
partial. China-aligned topic censorship appears on politically sensitive prompts. Safety coverage
is lighter than at frontier labs, with no companion guard model. A 1T-parameter model carries an
operational burden: multi-node infrastructure is required, with no small fallback variant.
How this scores
The ownership factor this domain covers, drawn from the one entry record.
2
TransparencyDo you know what it is: weights, training, behaviour, and legible terms?
WeakThe model cannot be inspected: the technical report documents a 1T/32B MoE trained on 15.5T tokens with MuonClip, but the training corpus and training code are closed, and the weights carry China-aligned topic censorship that cannot be inspected.
How this scores (AOI sub-dimensions)
Provenance3/5how well we can trace and verify what went into the modelDistributed from the verified moonshotai org on Hugging Face as block-FP8 safetensors with checksums, with no canonical-org malicious incident on record (checklist 5/8).
Governance3/5how accountable and well-documented the publisher isActive, accountable publisher (Moonshot AI) with a detailed technical report and a verified hub presence.
What this means for adoptionYou substantially use, modify and commercialise the self-hosted Kimi K2 weights under a near-MIT grant. The sole condition is the 100M-MAU / $20M-revenue trigger to display 'Kimi K2' in your UI. The weights run entirely on your own infrastructure, so your data stays yours. Ownership is partial because of transparency: the 15.5T-token corpus and training code are closed and the model carries China-aligned censorship that cannot be inspected. The weights and the hosted service differ: the platform.kimi.ai API trains on your content by default (opt-out only by enterprise agreement, Singapore law), so self-host if data control matters. Confirm the attribution trigger and the per-checkpoint LICENSE before shipping, and budget multi-node infrastructure to run it at all.
Sources
The same evidence records as the entry sheet. Read means the text was verified; unverified means it is known to exist but not yet read.
Model cardread2026-11-19
The Kimi-K2-Instruct model card on the verified moonshotai Hugging Face org describes block-FP8 safetensors, 1T total and 32B active parameters, a 128K context and the Muon optimiser.
Licenceread2026-11-19
The Kimi K2 LICENSE is standard MIT with one added clause, which opens "Our only modification part is that".
Documentationunverified2026-07-25
The moonshotai/Kimi-K2 repository documents the MoE architecture (384 experts, 8+1 selected), the MuonClip optimiser, block-FP8 weights and deployment on vLLM, SGLang, KTransformers and TensorRT-LLM with an OpenAI/Anthropic-compatible API.
Technical_reportread2026-07-25
Moonshot's arXiv technical report "Kimi K2: Open Agentic Intelligence" (2507.20534) describes a 1T-total / 32B-active MoE trained on 15.5T tokens with the MuonClip optimiser (Muon + QK-clip) and reports agentic and coding results (e.g.
Terms of serviceread2026-11-19
The platform.kimi.ai model-use agreement (page dated "Last Updated: July 30th, 2026") says user content is used to improve the services, with opt-out available only via an enterprise or separate written agreement.
Privacy Policyread2026-11-19
The platform.kimi.ai privacy policy says the hosted service trains on user prompts, audio, images, videos and files by default; the controller is MOONSHOT AI PTE.
Documentationread2026-11-19
The hosted Kimi API Platform notice says that from 2026-10-31 K2 series models are no longer available to newly registered users (existing users unaffected) and Moonshot plans to retire all K2 series models by the end of December 2026.
Third-party analysisunverified2026-07-25
On independent evaluation Kimi K2 is among the strongest open-weight models on agentic and coding benchmarks (e.g.
Third-party analysisunverified2026-07-25
Independent analysis notes Kimi K2, like other China-based open-weight models, applies China-aligned content filtering on politically sensitive topics.
Third-party analysisunverified2026-07-25
Kimi K2 Instruct is safety-tuned but with lighter alignment coverage than Western frontier labs and no companion guard model.
Third-party analysisunverified2026-07-25
No public EU AI Act training-content summary, copyright policy, or provider documentation package is published for Kimi K2, and the training corpus is not released.
Third-party analysisunverified2026-07-25
Kimi K2 is served across vLLM, SGLang, KTransformers and TensorRT-LLM, but its 1T-parameter scale requires multi-GPU / multi-node infrastructure even in block-FP8.