Kimi K2.7
#41 in Frontier LLMsMoonshot Ai · since 12. Juni 2026 · 22× · last seen Jul 13, 2026
40
Momentum
Kimi K2.7 (officially "Kimi K2.7 Code") is an open-weight language model developed by Moonshot AI using a Mixture-of-Experts architecture (1 trillion total parameters, 32 billion active parameters per token). Released on June 12, 2026 as a coding- and agent-focused successor to K2.6, it supports a 256K token context window (262,144 tokens) and is natively multimodal (text, image, video). The model always runs in mandatory "thinking" mode, while the company reports roughly 30% lower reasoning-token usage versus K2.6, and the weights are available under a Modified MIT License on Hugging Face.
Momentum trend
21.04.20.07.
Features
| Key Benchmark (%) | MCP Mark Verified: 81.1% (vs. Claude Opus 4.8: 76.4%); Kimi Code Bench v2: 62.0 (K2.6: 50.9) |
| Context Window (Tokens) | 256K tokens (262,144 tokens) |
| License | Modified MIT License (model and code repository) |
| Multimodality | Natively multimodal: text, image, and video input (via MoonViT encoder with 400M parameters) |
| Platform | Kimi API (platform.moonshot.ai), Kimi Code CLI, Hugging Face, ModelScope, OpenRouter, Cloudflare Workers AI, GitHub Copilot |
| Preis pro 1M Input-Tokens | $0,95 pro 1M Input-Tokens (Cache Miss) / $0,19 pro 1M Input-Tokens (Cache Hit) – offizieller API-Preis auf platform.kimi.ai |
| Price per 1M Tokens | Input: $0.95 (cache miss) / $0.19 (cache hit); Output: $4.00 (official Kimi API pricing) |
| Release Date | June 12, 2026 |