Windows settings backup becoming a new resilience baseline
July 7, 2026Teamcenter on Oracle AI Database@Azure: Architecture, Validation, and Results
July 7, 2026
Modern software engineering rarely happens in a single step, developers must refactor across large codebases, implement features spanning multiple files, and debug issues over long-running sessions. Kimi K2.7 Code from Moonshot AI, available on Microsoft Foundry starting July 1, 2026 in public preview, is designed to tackle these challenges. Building on the K2.6 series, it delivers meaningful advances in end-to-end task success, reasoning efficiency, and developer productivity scenarios, making it a more effective tool for real-world, long-horizon coding workflows.
With this launch, developers can access Kimi K2.7 Code through the Foundry model catalog and integrate it seamlessly into enterprise-grade applications.
What’s New: Advancing Beyond K2.6
Kimi K2.7 Code represents a focused evolution of the K2 family, with improvements designed specifically for real-world coding scenarios:
Long-Horizon Coding & Multi-Step Execution
- Designed to carry tasks across extended workflows and multiple steps
- More reliable instruction-following across long contexts
- Supports full engineering tasks from planning → execution → completion
Stronger End-to-End Task Completion
- Improves success rates on complex software engineering tasks
- Designed to complete tasks, not just generate individual code snippets
Agentic Task Execution
- Evaluated on benchmarks for autonomous agent workflows
- Supports systems where the model:
- reasons
- acts
- iterates across multiple steps
Optimized Reasoning Efficiency
- According to Moonshot AI, Kimi K2.7 Code reduces “thinking-token” usage by approximately 30% vs K2.6
- Achieves higher benchmark performance while consuming fewer tokens
- Enables:
- faster responses
- lower inference cost
- more work per context window
Benchmark Performance
Kimi K2.7 Code delivers substantial gains over K2.6 across both coding and agentic benchmarks:
|
Benchmark |
Kimi K2.6 |
Kimi K2.7 Code |
GPT-5.5 |
Claude Opus 4.8 |
|
Coding |
||||
|
Kimi Code Bench v21 |
50.9 |
62.0 |
69.0 |
67.4 |
|
48.3 |
53.6 |
69.1 |
63.8 |
|
|
MLS Bench Lite |
26.7 |
35.1 |
35.5 |
42.8 |
|
Agentic |
||||
|
Kimi Claw 24/7 Bench2 |
42.9 |
46.9 |
52.8 |
50.4 |
|
69.4 |
76.0 |
79.4 |
81.3 |
|
|
MCP Mark Verified3 |
72.8 |
81.1 |
92.9 |
76.4 |
1Kimi Code Bench v2 is an in-house benchmark developed by Moonshot AI
2Kimi Claw 24/7 Bench is an in-house benchmark for agentic evaluation for Moonshot AI
3MCP Mark Verified is a human-verified edition of MCPMark
Pricing and deployment
|
Model Name |
Deployment Type |
Input/1M tokens |
Output/1M tokens |
Cached Input/1M tokens |
|
Kimi K2.7 Code |
Global Standard
|
$0.95 |
$4.00 |
$0.19 |
Why Run Kimi K2.7 Code on Microsoft Foundry
Running Kimi K2.7 Code on Foundry allows developers to move from experimentation to production faster and more securely. Foundry offers:
Enterprise-Ready Security & Compliance
- Built-in governance, data protection, and responsible AI tooling
- Enterprise-grade deployment aligned with Microsoft compliance standards
Deep Developer Tooling Integration
- Native compatibility with:
- GitHub Copilot workflows
- VS Code and API-based pipelines
Unified Model Catalog
- Access Kimi alongside a growing ecosystem of frontier and open models
- Easily compare, evaluate, and switch models as needs evolve
Faster Time to Production
- Pre-integrated infrastructure eliminates setup complexity
- Streamlined path from experimentation → deployment
Get started today
Kimi K2.7 Code is now available in Foundry model catalog. Read more about the model details and try it out today. Start evaluating with your own datasets, and building and deploying applications in minutes.