Kimi K2.8 Preview API - 1M Context Coding Agent

What Kimi K2.8 Preview is

Kimi K2.8 Preview is Moonshot AI's coding-focused model that became the default engine in Kimi Code on September 11, 2026. It is a silent upgrade to the default Kimi Code coding model rather than a standalone open-weight release, with a 1M-token context window, adjustable thinking, and multimodal input.

Where it fits

It is built for coding workflows where long context, adjustable thinking, and agentic execution matter.

• Large-codebase refactoring: trace dependencies across files, apply coordinated edits, and run tests to verify the change.
• Code completion: turn repository context and precise requirements into implementation-ready functions, modules, and fixes.
• Coding Agents: plan multi-step tasks, call tools, inspect results, and continue iterating until the task is complete.
• Screenshot and screen-recording to code: use image and video input to reproduce interfaces, diagnose visual defects, and refine rendered output.

Architecture and Agent capabilities

Kimi K2.8 Preview is a hosted coding-model upgrade delivered through Kimi Code and enabled API gateways. This page does not provide a downloadable weight package; confirm deployment and access details in the official Moonshot AI documentation.

• Native multimodal: It supports text, image, and video input for vision-in-the-loop work.
• Reasoning: Thinking can be adjusted with reasoning_effort values low, high, and max. Turning thinking off routes requests to the non-thinking version of Kimi K2.8 Preview.
• Tools: Function calling, tool_choice, and JSON Schema structured output support Agent integrations.
• Access model: Kimi K2.8 Preview is a silent upgrade to the default Kimi Code coding model, not a standalone downloadable open-weight release.

Kimi K2.8 Preview vs Kimi K3 vs K2.7 Code HighSpeed

All three models serve the Kimi Code family. Choose based on capability, context, reasoning depth, and output speed.

• Choose Kimi K3 when you need the strongest model for large projects, deep reasoning, and complex long-context Agent tasks, and can accept higher membership requirements and usage.
• Choose Kimi K2.8 Preview for daily development and value: it is reported close to K3, more thinking-efficient, and makes 1M context available across all membership tiers.
• Choose Kimi K2.7 Code HighSpeed only when output speed is the priority; it is the previous-generation option and is otherwise surpassed by K2.8 Preview in context, thinking efficiency, and overall capability.

Integrate Kimi K2.8 for coding Agents

Load only the repository files and cross-file dependencies needed for the task into the 1M-token window. Then let the Agent run tests, inspect diffs, and verify that the change actually works.

curl https://tokenra.io/zen/go/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"kimi-k2.8-preview","messages":[{"role":"user","content":"Refactor the attached module to remove duplicated state, run the test suite, and report which tests fail with the likely cause."}],"reasoning_effort":"high"}'

Pricing

Kimi K2.8 Preview is a paid model through TokenRa. Current rates are $0.80 per 1M input tokens, $3.35 per 1M output tokens, and $0.14 per 1M cached read tokens. Confirm current availability and account limits in the live console.

Production checklist

• Keep API keys on the server and separate credentials by environment or workload.
• Set explicit timeouts, bounded retries, and request tracing before routing user traffic.
• Evaluate model behavior with representative inputs instead of assuming versions are interchangeable.
• Monitor usage and cost in the console, then add application-side budgets and limits.
• Review provider terms, privacy requirements, and output rights for your use case.

API FAQ

Continue exploring