On June 12, 2026, Moonshot AI published Kimi K2.7-Code on Hugging Face under a Modified MIT license, marking a significant milestone in democratized access to high-performance AI. This Mixture-of-Experts model with 1 trillion total parameters — only 32 billion active per token across 384 specialized experts — outperforms Anthropic's Claude Opus 4.8 on tool-use benchmarks at a fraction of the price. With a 256K-token context window and an API starting at $0.95 per million input tokens, Kimi K2.7-Code removes the economic barrier that has prevented small and medium businesses from accessing enterprise-grade code automation until now.
What Did Moonshot AI Release on June 12?
Moonshot AI released two products simultaneously: Kimi K2.7-Code (the specialized language model) and Kimi Work (a local desktop agent with a swarm of up to 300 sub-agents). Kimi K2.7-Code powers both. Technically, it is a 1T-parameter MoE architecture with 32B active parameters per token and 384 experts, designed specifically for long-horizon agentic software engineering. Published benchmarks show: +21.8% on Kimi Code Bench v2, +11.0% on Program Bench, +31.5% on MLS Bench Lite, and approximately 30% fewer reasoning tokens than Kimi K2.6 on equivalent tasks. On tool use — the critical capability for autonomous agents integrating with external APIs — it surpasses Claude Opus 4.8. The API is priced at $0.95/M input tokens and $4.00/M output tokens (model ID: kimi-k2.7-code). The full model is available on Hugging Face with open weights, also enabling local deployment. Kimi Code, the terminal-first coding agent built on K2.7-Code, starts at $19/month.
"An open-source 1-trillion-parameter model that outperforms closed first-tier solutions on code tasks isn't just a technical achievement — it's a structural shift in the economics of automation for SMBs."
Davarion Group & LabsReal Impact for SMBs
- 01Code automation at minimal cost: at $0.95/M input tokens, an SMB can generate thousands of scripts, process automations, or API connectors for under $10/month — no dependency on premium closed-model pricing.
- 02Superior tool use performance: K2.7-Code beats Claude Opus 4.8 on tool use benchmarks, making it the ideal choice for autonomous agents that must interact with external APIs, CRMs, ERP systems, or e-commerce platforms.
- 03Local deployment available (open weights): businesses with strict data privacy requirements — clinics, law firms, fintech companies — can run the model on their own servers without sending data to third parties.
- 04256K-token context window: enables processing entire codebases, extensive technical documentation, or long customer support email chains continuously, without losing context between requests.
- 05Immediate action available: the model is live today on Hugging Face and the Kimi API — any business can begin experimenting with code automation integrated into existing workflows within hours.
The release of Kimi K2.7-Code concretely reshapes the business automation equation: until now, first-tier agentic coding capabilities — autonomous code generation, debugging, and external API integration — were reserved for companies with six-figure annual AI budgets or advanced technical teams. With an open-weight model that outperforms Claude Opus 4.8 on the metrics most critical for autonomous agents, priced under $1 per million input tokens, SMBs can now build robust code automations: connectors between billing systems and online stores, agents that monitor and auto-fix production errors, and self-updating data analysis pipelines. The 30% reduction in reasoning tokens versus K2.6 also means lower operational costs in production, improving the ROI of every automated workflow.
At Davarion Group & Labs we build autonomous agents for SMBs in Houston, TX and throughout Latin America — and models like Kimi K2.7-Code are exactly the infrastructure that makes our work viable for businesses of any size. If your company needs to automate processes involving code, APIs, system integrations, or repetitive technical tasks, this is the ideal moment to start. Visit davarion.com and we'll show you how to deploy an autonomous coding agent calibrated with Kimi K2.7-Code for your specific business in days, not months.
