{"rewrite":{"id":"r_9a3ad9f4c3808d4772d73125","clusterId":"c_a8ec5bc24885f6e7a8ac9871","slug":"moonshot-ai-releases-kimi-k2-7-code-as-open-model","model":"deepseek-v4-flash","headline":"Moonshot AI Releases Kimi K2.7 Code as Open Model","summary":"Chinese AI company Moonshot AI released Kimi K2.7 Code on June 12, 2026, an open model specialized for coding agents. The MoE model has 1 trillion total parameters and 32 billion active parameters with a 256K token context. It shows improved coding performance over Kimi K2.6 and uses about 30% fewer thinking tokens per task.","whyItMatters":"Kimi K2.7 Code is the latest open coding model from Moonshot AI, offering better performance and lower token consumption than its predecessor, which matters for developers seeking efficient agentic coding tools.","webCardHtml":"\u003cp\u003eMoonshot AI released Kimi K2.7 Code on June 12, 2026, as an open model available for download on Hugging Face. The model is designed for coding agents and uses a Mixture of Experts architecture with 1 trillion total parameters and 32 billion active parameters. Its maximum context length is 262,144 tokens. Benchmark results show Kimi K2.7 Code outperforms the previous generation Kimi K2.6 on coding tasks while consuming about 30% fewer thinking tokens per task, addressing the problem of wasteful token accumulation in AI agents.\u003c/p\u003e","blueskyPost":"Moonshot AI's Kimi K2.7 Code uses 30% fewer thinking tokens per task than its predecessor while improving coding performance. The efficiency gain matters more than the parameter count for production deployment.","twitterPost":"Kimi K2.7 Code cuts thinking tokens 30% while improving coding performance. Efficiency, not scale, is the headline here.","threadsPost":"Moonshot AI released Kimi K2.7 Code as an open model on June 12. It beats Kimi K2.6 on coding benchmarks while using roughly 30% fewer thinking tokens per task. For developers running agents at scale, the token savings may matter more than the parameter total.","newsletterBlurb":"Moonshot AI released Kimi K2.7 Code, an open model specialized for coding agents. The MoE model has 1 trillion total parameters and uses about 30% fewer thinking tokens than Kimi K2.6 while improving coding performance.","attributionJson":"[{\"source\":\"GIGAZINE\",\"url\":\"https://gigazine.net/news/20260615-kimi-k2-7-code/\",\"title\":\"Chinese-Made AI Model 'Kimi K2.7 Code' Released as Open Model, Best Coding Performance in Kimi Series with Lower Token Consumption\"}]","lintFlagsJson":null,"lintHits":0,"costUsd":0,"inputTokens":4386,"outputTokens":532,"status":"published","repairAttempts":0,"nextRepairAt":null,"factsAttemptedAt":1781491321,"createdAt":"2026-06-15T02:30:35.000Z","publishedAt":"2026-06-15T02:34:57.000Z","updatedAt":"2026-06-15T02:34:57.000Z"},"cluster":{"id":"c_a8ec5bc24885f6e7a8ac9871","canonicalTitle":"中国製AIモデル「Kimi K2.7 Code」がオープンモデルとして公開される、Kimiシリーズ最高のコーディング性能でトークン消費も少なめ","representativeArticleId":"a_dc6ed9c0ffe7900c6233d4d1","sourceCount":1,"writtenSourceCount":1,"writeAttempts":0,"isSolo":true,"entitiesJson":"{\"anime_titles\":[],\"manga_titles\":[],\"work_titles\":[],\"studios\":[],\"people\":[],\"type\":\"news\",\"domain\":\"other\",\"is_roundup\":false}","contentType":"news","status":"published","firstSeenAt":"2026-06-15T01:45:00.000Z","lastSeenAt":"2026-06-15T01:45:00.000Z","updatedAt":"2026-06-15T02:34:57.000Z"},"attribution":[{"source":"GIGAZINE","url":"https://gigazine.net/news/20260615-kimi-k2-7-code/","title":"中国製AIモデル「Kimi K2.7 Code」がオープンモデルとして公開される、Kimiシリーズ最高のコーディング性能でトークン消費も少なめ"}],"entities":{"anime_titles":[],"manga_titles":[],"work_titles":[],"studios":[],"people":[],"type":"news","domain":"other","is_roundup":false},"keyFacts":["Moonshot AI released Kimi K2.7 Code on June 12, 2026, as an open model on Hugging Face.","The MoE model has 1 trillion total parameters and 32 billion active parameters with a 256K token context.","Kimi K2.7 Code outperforms Kimi K2.6 on coding benchmarks while using about 30% fewer thinking tokens per task."]}