Open weightsMoonshot AI/10 trustVerified

Kimi K2.7 Code

Name: Kimi K2.7 Code
Rating: 8

Moonshot's open-weights agentic coding model — 1T-param MoE tuned for long-horizon software engineering

Visit official page →

About

Kimi K2.7 Code is Moonshot AI's coding-specialized open-weights model, released June 12, 2026 — the latest release in the K2 line (K3 is a separate flagship line). It keeps the K2.6 MoE backbone (1 trillion total parameters, 32B active per token, 384 experts) with training tilted toward coding data, forced always-on thinking mode, a 256K-token context window, and the MoonViT vision encoder for image and video input. Weights are published on Hugging Face under a Modified MIT license, with vLLM/SGLang/KTransformers support and an OpenAI-compatible API.

Specs

Context window

256K tokens (262,144)

Pricing

$0.95 input / $4.00 output per 1M tokens ($0.19 cached input, Moonshot API)

Free quota

No free API tier published (API key is free; a $1 minimum prepaid balance activates calls per a Sept 2026 guide); weights are free to download and self-host

API access

OpenAI-compatible API at api.moonshot.ai (model id kimi-k2.7-code); also on OpenRouter and other third-party hosts; Kimi Code CLI available

Weights

Open weights

Released

June 12, 2026

Benchmarks

Kimi Code Bench v2: 62.0 vs K2.6 50.9 (+21.8%) — provider-reported (Moonshot's own benchmark)
Program Bench: 53.6 (+11.0% vs K2.6) — provider-reported
MLS Bench Lite: 35.1 (+31.5% vs K2.6) — provider-reported
MCP Mark Verified: 81.1 vs Claude Opus 4.8 76.4 — provider-reported (Moonshot's own table; harness differences apply)
Artificial Analysis Intelligence Index: 26, above the open-weight median of 18 — independent
Artificial Analysis measured K2.7-Code above Claude Sonnet 4.6 (non-reasoning) on the Intelligence Index — independent (via Totalum, July 2026)
~30% fewer thinking tokens per task vs K2.6 — provider-reported

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

codeautomationreasoningchatvisiontesting

Trust score — 8/10

/10 trustVerified

Release, HF repo, license, context, pricing and AA independent scores all verified against multiple sources (HF model card via MarkTechPost, Artificial Analysis, spectrumailab license table). Minor caveats: headline benchmarks are Moonshot's own tables; Modified MIT requires visible credit above ~100M MAU / $20M monthly revenue. Metrics-v1 score 8/10 — adoption 8.5, maintenance 8, openness 7.2, manual 9.

How the trust score is built

Full methodology →

Adoption · stars/downloads, log-scaled vs peers: 8.5/10
Maintenance · recency, releases, issue load: 8.0/10
Openness · license + docs: 7.2/10
Manual review · sanity read + red-flag check: 9.0/10

· HF top match: unsloth/Kimi-K2.7-Code-GGUF
· HF downloads 304,694 (peer-normalized)
· released 123 days ago
· manual review: release/HF/license/pricing/AA scores verified; headline benchmarks are Moonshot's own tables