Bring a large codebase, a document set or visual evidence into one conversation. Kimi K3.1 Pro combines a million-token context with reasoning and tool calling, so you can work through the details before accepting the answer.
Start with the prompt in the composer or choose an example.
Credit-based — 1 credit = $0.001 USD. Pay only for completed generations.
One 1K-token prompt with a 1K-token answer at the cheapest tier.
Kimi K3.1 Pro is Moonshot AI's reasoning model for coding and knowledge work, developed through further training of Kimi K3. Its 1,048,576-token context gives a task room for source material, conversation history and the answer. Kimi K3.1 Pro can read visual inputs and return text or function calls; your application supplies the tools and checks their results. Start Kimi K3.1 Pro with a real task you can verify, such as a failing test or a comparison of supplied documents.
Real-world workflows and production use cases you can build and ship with this model.
Give Kimi K3.1 Pro the relevant files, a failing test and the behavior you expect. Ask for a proposed change with tests, then return the test output for another pass. Kimi K3.1 Pro can request tools during that loop, while your application controls execution and decides when the change is ready to review.
See the request format
Use Kimi K3.1 Pro to compare a specification with its implementation notes, or reconcile claims across several reports. Ask it to identify the passage behind each conclusion and separate missing evidence from disagreement. The Kimi K3.1 Pro context keeps more of the supplied material together; check Kimi K3.1 Pro quotations and calculations against the originals.
Pair a screenshot with the behavior it should show, then ask Kimi K3.1 Pro to explain the discrepancy. For visual debugging, combine the image with component code and an explicit test. Moonshot also documents video understanding, but the current reAPI video check returned unsupported; use images for the verified visual workflow.
Use the playground to inspect an answer before building a larger workflow. For Kimi K3.1 Pro, compare the proposed fix, supporting evidence and total usage against a task you already understand. The pricing table separates input, output and cached input, so repeated context is visible in your cost planning.
The chat format fits an existing OpenAI-compatible client. Preserve the complete Kimi K3.1 Pro assistant message between turns, including reasoning and tool calls, so the next request carries the state the model expects. The documentation includes a tool-result round trip and examples in four languages.
Use plain text for an explanation, a JSON schema for extraction, or function calls when a tool should supply evidence. Kimi K3.1 Pro exposes these choices in the playground. A valid structure helps your application read the result; your own checks determine whether its contents are right.
Sign in and create an API key in your account settings. Keep it on your server when you connect Kimi K3.1 Pro to an application.
OpenChoose kimi-k3.1-pro as the model and include your messages. Start Kimi K3.1 Pro with a small, checkable task and an explicit reasoning effort.
OpenRead the answer and usage, then run your own checks. Keep the complete Kimi K3.1 Pro assistant message when you add a tool result or a follow-up question.
OpenKimi K3.1 Pro supports 1,048,576 tokens shared by input and output. Plan room for the answer when assembling a large request.
Moonshot documents a 131,072-token default output allowance for Kimi K3.1 Pro, with a higher ceiling within the shared context. See the docs for measured limit behavior.
Kimi K3.1 Pro always reasons. Choose low, high or max; max is the official default. Compare the settings using your own acceptance tests.
Kimi K3.1 Pro offers function calling and JSON Schema output. Execute requested tools in your application, then return their results for the next turn.
Kimi K3.1 Pro read the image fixture in integration checks. Public image URLs are prepared by reAPI; video is officially documented but currently returned unsupported.
Kimi K3.1 Pro supports prefix caching. Keep repeated material at the beginning, and check reported cache usage instead of assuming every repeated prompt hits.
Common questions about this model.
Both models can support a substantial coding or analysis task. Compare Kimi K3.1 Pro with Claude Opus 5.5 using the same evidence and a result you can test. These rows describe official model contracts; the linked guide explains why a benchmark result and a working application require different checks.
Official documentation checked October 24, 2026. Batch-only limits and Highspeed claims are outside this comparison. Consult the Kimi K3.1 Pro API docs for the dated behavior observed through reAPI.
Explore more models in the same category.
Moonshot AI
From $2.50 per 1M tokens
Anthropic
From $2.00 per 1M tokens
OpenAI
From $1.60 per 1M tokens
Anthropic
From $2.40 per 1M tokens
curl https://reapi.ai/api/v1/chat/completions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"kimi-k3.1-pro","messages":[{"role":"user","content":"Review this function for edge cases."}],"reasoning_effort":"low"}'Rejected synchronously with the failing field named. Check enum values, ranges, and URL fields — nothing is charged.
401 means a missing or invalid Bearer key; 402 means the reserve exceeds your balance. Manage keys and credits in the console.
Reference material and outputs pass automated moderation. Rejected tasks fail with a clear error and a full refund.
A task that reaches a failed state is never charged — the reserve refunds automatically. Keep the task id and retry when ready.
Try it in the playground or grab an API key to integrate now.