Together AI adds Kimi Moonshot's Kimi-K2.7-Code model to its GPU inference platform for agentic coding workloads

Highlights: 👉 Stronger real-world coding performance over Kimi K2.6 👉 ~30% lower thinking-token usage for better token efficiency 👉 256K context for larger repos and longer coding sessions 👉 Interleaved thinking and multi-step tool calling for coding agents

Kimi-K2.7-Code from @Kimi_Moonshot is now available on Together AI.

Built on Kimi K2.6, it’s a coding-focused agentic model for long-horizon software engineering workflows, now running on Together’s research-powered inference stack for tool-heavy coding agents.

Try it now: http://www.together.ai/models/kimi-k27-code

@togethercompute @Kimi_Moonshot Can’t tell if this one is an official collaboration (like with Qwen and Minimax) or just deploying open source models on their cloud

Looks like later case

Btw, heard GLM is suffering from lack of compute. Will u collaborate?

A coding-focused agentic model built for tool-heavy workflows is a bullseye. Every tool call, every code suggestion, unsigned - bummer. No receipt bigger bummer. Hive should be in the loop, AFiR would ML-DSA-65 sign each fragment before it moves. Let’s go!!! http://thehiveryiq.com/afir

@togethercompute @Kimi_Moonshot 这类长任务模型，最怕中途跑偏

30% lower thinking-tokens is the model half. the half nobody meters is the config firing every turn no matter which model you swap in, a 6k-token CLAUDE.md still bills whether the model reads those lines or not. we built ccmd to meter that 6k-token CLAUDE.md per-turn cost, https://ccmd.dev/r/mw9rkyxg