VidgoAI

Kimi K3 API

moonshotai/kimi-k3
1,048,576 tokens · 456 input credits / 1M tokens

Kimi K3 turns conversation messages and long text context into generated answers, plans, code, and document analysis. Its 1,048,576-token context is suited to repository-scale review, long document sets, and sustained agent workflows through Chat Completions.

Kimi K3

Moonshot AI · chat-completions

Chat with Kimi K3

Each model has its own conversation. Switching never sends another model's history, and switching back resumes where you left off. Requests are billed from actual token usage.

Ctrl / ⌘ + Enter to send0 / 32,000

Continue with

GPT 5.6 SolGPT 5.6 TerraGPT 5.6 Luna

Kimi K3

Kimi K3 accepts an ordered, non-empty messages array and returns generated text plus usage in synchronous JSON or SSE. Organize long repositories or document sets with explicit boundaries so the model can connect evidence across distant parts of the conversation.

Why Choose This?

  • 1,048,576-token contextPlace long documents, repository excerpts, and prior messages in one conversation when relationships span distant sections.

  • Long-horizon codingProvide goals, code context, and checkpoints for implementation planning and sustained software tasks.

  • Agent-oriented analysisDescribe available tools and completion criteria to generate stepwise plans for multi-stage work.

Parameters

ParameterRequirementDescription
messagesRequired

A non-empty ordered conversation array; each message provides a role and text content.

max_tokensOptional

Sets the maximum number of tokens the model may return when the selected deployment accepts the value.

streamOptional

Set false for synchronous JSON or true for an ordered SSE event stream.

Defaultfalsetrue

How to Use

  1. Organize the long contextGroup repository files, documents, and prior decisions into clearly labeled sections.

  2. Mark source boundariesName each file or document and preserve its section identifiers for traceable analysis.

  3. Define milestones and toolsDescribe the expected stages, available actions, and completion criteria for sustained agent work.

  4. Verify against the sourcesCheck cited sections, proposed code changes, and unresolved assumptions before continuing execution.

Pricing

Vidgo AI meters input, output, cache-read, and cache-write tokens separately, then settles their sum in credits. Rates below are per 1M tokens.

UsageRateDetails
Input tokens456 credits / 1M tokensInput usage that does not hit prompt cache.
Output tokens2280 credits / 1M tokensTokens generated by the model.
Cache read456 credits / 1M tokensInput tokens read from prompt cache.
Cache write456 credits / 1M tokensInput tokens written to prompt cache.

Best Use Cases

  • Repository-scale implementation briefCombine source files, architecture notes, and issue history into a scoped change plan with validation steps.

  • Multi-document research mapTurn reports and specifications into linked findings, source references, disagreements, and open questions.

  • Long-running agent planConvert a goal, tool inventory, and prior progress into ordered milestones and completion checks.

Pro Tips

  • Add stable file names, document titles, and section markers so answers can point back to specific long-context sources.
  • Ask for source identifiers beside important findings when the conversation contains many documents or repository excerpts.
  • Remove unrelated history before a new phase so the million-token window is reserved for evidence that affects the current task.

Notes

  • The context window is 1,048,576 tokens; the maximum output depends on the selected deployment.
  • Kimi K3 uses /v1/chat/completions on Vidgo AI.
  • The minimum positive charge for a successful request is 0.001 credit; requests without valid output are not charged.
  • If a provider returns valid content without usage, the backend estimates token usage for settlement.

Related Models

GPT 5.6 SolOpenAI · 1M tokensGPT 5.6 Sol text generation API from 280 credits per 1M input tokens.
GPT 5.6 TerraOpenAI · 1M tokensGPT 5.6 Terra text generation API from 112 credits per 1M input tokens.
GPT 5.6 LunaOpenAI · 1M tokensGPT 5.6 Luna text generation API from 11.2 credits per 1M input tokens.
Claude Opus 5Anthropic · 1M tokensClaude Opus 5 text generation API from 400 credits per 1M input tokens.
Claude Sonnet 5Anthropic · 1M tokensClaude Sonnet 5 text generation API from 170 credits per 1M input tokens.

Kimi K3 API — Frequently asked questions

What is the Kimi K3 API?

Kimi K3 is developed by Moonshot AI. It accepts ordered conversation messages and returns generated text plus usage through Vidgo AI.

How do I call the Kimi K3 API?

Send a Bearer-authenticated POST request to /v1/chat/completions with model kimi-k3 and a non-empty messages array. Use stream: false for JSON or stream: true for SSE.

How much does Kimi K3 cost?

Per 1M tokens, Vidgo charges 456 credits for input, 2280 for output, 456 for cache reads, and 456 for cache writes. The minimum positive charge for a successful request is 0.001 credit.

What inputs does Kimi K3 accept?

The Vidgo Chat Completions endpoint accepts a non-empty messages array with role and text content. It also accepts max_tokens and the stream switch documented for this protocol.

How large is the Kimi K3 context window?

Kimi K3 has a 1,048,576-token context window. Use clear file, document, and section labels when a request spans a large repository or document set.