Open model Kimi K2.7 Code undercuts GPT-5.5 and Claude by up to 12x on price per token

2026-06-13 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Robotics & Autonomous Systems · Depth: Intermediate, short

Summary

Moonshot AI released Kimi K2.7 Code on June 13, 2026, an open-source, multimodal AI model specifically engineered for complex programming tasks and agent-based workflows. Available on Hugging Face, this Mixture-of-Experts model features one trillion total parameters with 32 billion active per token and a 256,000 token context length. While K2.7 Code trails Western competitors like GPT-5.5 and Claude Opus 4.8 on standard coding benchmarks (e.g., 53.6 on Program Bench compared to GPT-5.5's 69.1), it demonstrates strong performance in agent-oriented tests, even surpassing Claude Opus 4.8 on MCPMark Verified with a score of 81.1. Crucially, its API pricing is \$0.95 per million input tokens and \$4.00 per million output tokens, making it up to 12 times more cost-effective than models like Claude Fable 5. The model also includes a native INT4 quantization option and a modified MIT license for commercial use.

Key takeaway

For AI Engineers or ML Directors evaluating coding models for agent-based systems, you should benchmark Kimi K2.7 Code against your specific tasks. Its significantly lower API cost—up to 12x cheaper than alternatives like Claude Fable 5—means you can achieve substantial savings, even if raw performance slightly trails top-tier models. Prioritize task-specific evaluations to determine if its "good enough" performance at a fraction of the cost makes it your optimal choice for high-volume agentic workflows.

Key insights

Kimi K2.7 Code offers a cost-effective, agent-optimized open-source model for complex coding, despite trailing top benchmarks.

Principles

• Cost-effectiveness can outweigh raw benchmark scores.
• Agentic benchmarks reveal practical model strengths.
• MoE architectures enable large parameter counts with efficient inference.

Method

The model uses a Mixture-of-Experts (MoE) architecture with 384 experts, activating 8 per token from one trillion total parameters, and employs "preserve_thinking" mode for agentic scenarios.

In practice

• Evaluate K2.7 Code for agent-based coding workflows.
• Utilize INT4 quantization for hardware efficiency.
• Compare cost-per-token against raw performance for specific tasks.

Topics

• Kimi K2.7 Code
• Open-source Models
• Agentic AI
• Code Generation
• API Pricing
• Mixture-of-Experts

Best for: CTO, VP of Engineering/Data, AI Architect, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• The Dragon’s Code vs The Anthropic Giant: How Kimi K2.5, — New Chinese open-source models are challenging Claude Opus 4.6's dominance in AI coding agents.
• Kimi K2.6: BEST Opensource AI Model That Beats Opus 4.6 and Gemini 3.1 Pro (Fully Tested) — Kimi K2.6 is a cost-efficient, open-source coding model with "state-of-the-art" long-horizon execution and multi-agent capabilities.
• Claude Code Heading for a $100 Paywall Instead of $20! The New Pricing Reality for AI Coding Agents — Advanced AI coding agent services are shifting from flat-rate, low-cost subscriptions to higher-priced tiers due to increased compute demands.
• How to Run OpenClaw with Open-Source Models — Cost-effective open-source LLMs like Kimi-K2.5 can power OpenClaw effectively, despite minor performance trade-offs.
• OpenAI GPT-5.5: BEST AI Model Ever! Beats Opus 4.7 & Gemini 3.1! Powerful & Fast! (Fully Tested) — GPT 5.5 excels in multi-step tasks, coding, and efficiency, offering significant cost and performance advantages.
• Codex is gaining steam — AI platforms are evolving to be more accessible to non-technical users and offer competitive performance-to-cost ratios.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.