Moonshot AI announced Kimi K2.7 Code on its developer forum on June 16, 2026, as a coding-focused addition to the Kimi K2 family built on a Mixture-of-Experts architecture with roughly 1 trillion total parameters and 32 billion active parameters per token. The release notes lock sampling parameters to temperature 1.0, The announcement reports benchmark improvements over K2.6 including Kimi Code Bench V2 at 62.0 (vs 50.9), Program Bench at 53.6 (vs 48.3), MLS Bench Lite at 35.1 (vs 26.7), Kimi Claw 24/7 Bench at 46.9 (vs 42.9), MCP Atlas at 76.0 (vs 69.4), and MCP Mark Verified at 81.1 (vs 72.8). New capabilities include multimodal t

GitHub's official changelog announces Kimi K2.7 Code as generally available in GitHub Copilot on July 1, 2026, describing it as "the first open-weight model offered as a selectable option in the Copilot model picker" and hosted by GitHub on Microsoft Azure, billed at provider list pricing under usage-based billing. Ini The model is selectable in the Copilot model picker across VS Code 1.127.0+, Visual Studio 17.14.6+, Copilot CLI, Copilot cloud agent, the Copilot App, github.com, GitHub Mobile (iOS/Android), JetBrains 1.9.1-251+, Xcode, and Eclipse. For Copilot Business and Enterprise, K2.7 Code is off by default and requires plan ad

CoreWeave's official blog (June 18, 2026) announces Kimi K2.7 Code availability on CoreWeave Serverless Inference and claims the top position on Artificial Analysis's speed-vs-price chart for the model, citing metal-to-model optimization and collaboration between its Applied Training and Inference teams. The post repea The engineering details highlighted for the CoreWeave deployment include NVFP4 quantization targeting NVIDIA Blackwell GPUs and a DFlash speculator to boost throughput. CoreWeave frames the optimization as a continuation of its prior K2.6 work and positions the result as relevant to production inference workloads where

Independent benchmark aggregator LLM Stats profiles Kimi K2.7 Code with a composite LLM Stats Score ranking it 56th overall, capability tiers of Tool Calling 34 of 183, Coding 47 of 257, Reasoning 81 of 349, and Math 167 of 317, and a Quality Tracker trend of +1.10σ marked Stable based on 97 votes over the prior week a Per-benchmark scores for Kimi K2.7 Code on the aggregator include MCP-Mark at 0.81/1 (rank 1) under a verified 100-step tool-call budget and 32k max tokens per step averaged over 3 runs, MCP Atlas at 0.76/1 (rank 13) using the official configuration with a 100 tool-call budget, and LiveBench at 0.72/1 (rank 27) using t

Cloudflare's official changelog confirms Workers AI launched `@cf/moonshotai/kimi-k2.7-code` on June 12, 2026. The model is described as a code-optimized MoE with 1T total parameters and 32B active per token, delivering benchmark gains over K2.6 of +21.8% on Kimi Code Bench v2, +11.0% on Program Bench, and +31.5% on ML Capabilities listed include a 262.1k-token context window, vision inputs for images alongside text, a thinking mode with configurable reasoning depth via chat-template kwargs, multi-turn tool calling, and structured outputs with JSON schema support. For K2.6 migrators, the only API change is cached input pricing at $0.

OpenRouter's model page lists `moonshotai/kimi-k2.7-code` with text and image modalities, a 256K context, thinking mode always on, and a release date of June 12, 2026, at a base price of $0.66 input / $3.40 output per 1M tokens. The aggregator surfaces 14 hosting providers with varying pricing, latency, and throughput: Weighted-average prices customers actually pay are $0.2518/M input and $3.715/M output, reflecting caching and discounts below list rates. Cache-read pricing varies widely by provider (e.g., DeepInfra $0.136/M, Moonshot AI and Cloudflare $0.19/M, Moonshot AI Highspeed $0.38/M), giving developers concrete options to tra

Kimi's official Code docs detail three recent CLI releases relevant to K2.7 Code workflows: v0.41.0 (Sep 4, 2026) adds an experimental Tower multi-agent mode in Web, text-selection annotations across messages/diffs/terminals, and removes blocking of dangerous commands in auto permission mode; v0.40.0 (Sep 2, 2026) rena Also in these releases: a new session rating prompt (disableable in config), background-question answers routed directly to the agent instead of an output file, the `kimi session list` CLI command, and v0.40.1 fixing a reappearing migration prompt. While the page documents Kimi Code tooling rather than model benchmarks