Kimi K2.7 Code (NovitaAI)

Moonshot AI released Kimi K2.7 Code as a 1-trillion-parameter open-source coding model, with the article stating a June 18, 2026 launch and a 128,000-token context window under a modified Apache 2.0 license, per the supplied excerpt from freeainews.com. The model targets code generation, debugging, test writing, and re The article reports vendor-claimed benchmark results including 72.4 percent on SWE-bench Verified and 68.1 percent on LiveCodeBench, which the excerpt says put Kimi K2.7 Code ahead of DeepSeek V4 and Qwen3-Coder in vendor-run tests, with claims of beating GPT-5.1 Codex on HumanEval and matching Claude Opus 4.8 on SWE-b

The felloai.com review states Kimi K2.7 Code was released on June 12, 2026, by Moonshot AI as an open-weight, coding-focused model with 1 trillion total parameters, 32B active per token across 384 experts, and a 256K-token context window, shipping under a Modified MIT license on Hugging Face, per the supplied excerpt. According to the same excerpt, Moonshot claims roughly 30 percent fewer reasoning tokens than the previous K2.6 while scoring higher on its own coding benchmarks, with all benchmark gains being company-reported and no independent SWE-bench scores yet available. API pricing is listed at $0.95 per 1M input tokens and $4.

The thesys.dev blog describes Kimi K2.7 Code as Moonshot AI's open coding model built for agentic software engineering, improving on K2.6 across published coding and agentic benchmarks while using roughly 30 percent fewer thinking tokens, per the supplied excerpt. It retains the same 256K context window and is priced b Technical specifications from the excerpt include 1 trillion total (MoE), 32B active parameters per token, a 262,144-token (256K) context window, Multi-head Latent Attention (MLA), and a MoonViT vision encoder at 400M parameters, with text, image, and video input modalities, and open weights available on HuggingFace at