Kimi K3.1 Pro API Guide

Kimi K3.1 Pro API: 2.8T MoE, 1M context, lower price

Kimi K3.1 Pro is Moonshot AI's October 2026 refresh of Kimi K3, released on 23 October 2026. It keeps the 1,048,576-token context window and the sparse Mixture-of-Experts footprint of the previous generation — 2.8T total parameters with roughly 104B active per token, marketed as a 3T-class model — and cuts list pricing to $2.00 per 1M input tokens and $10.00 per 1M output tokens. On TokenRa the same model is billed at $1.60 and $8.00.

Kimi K3.1 Pro pricing on TokenRa

TokenRa bills per 1M tokens, counted separately for input, output and cached input reads. There is no platform subscription. The comparison column is Moonshot AI's published list price, not a competing gateway's rate.

Worked examples at these rates: a 400K-token document costs $0.64 to read in ($0.80 at list price); 10M input tokens plus 2M output tokens costs $32.00 ($40.00 at list price); filling the entire 1,048,576-token window once costs $1.68. A 1M-token prompt that is 90% cache hits costs about $0.30 instead of $1.60. Live per-account rates: /pricing | Check your access: /dashboard/overview

What changed from Kimi K3

Kimi K3.1 Pro is a cheaper, later refresh rather than a different architecture. Capability claims below are Moonshot AI's own figures, so treat them as claims to re-test on your inputs rather than guarantees.

Moonshot AI reports Terminal-Bench 4.0 at 65.7% and a 65% lower hallucination rate than Kimi K3 on its own internal evaluations. Both numbers come from the vendor's release notes, so benchmark them against your own prompts before you make a migration decision.

Integrate Kimi K3.1 Pro on TokenRa

TokenRa exposes an OpenAI-compatible chat endpoint. Copy the enabled model identifier from the console, keep document assembly, retrieval and citation checks in your application, and cap output length so the output rate does not dominate your bill.

Two identifiers are enabled for this model on TokenRa. Both run the same Kimi K3.1 Pro weights and the same 1,048,576-token context; they differ only in the rate applied to your account:

The -0.2x suffix marks a promotional channel that bills at 20% of Moonshot AI's list price — 2折 — which is a quarter of the standard TokenRa rate for the same model. Latency, context length and output quality are unchanged; only the price differs. Use it for bulk ingestion, evaluation runs and long-document processing where the total token bill dominates. At these rates a 400K-token document costs $0.16 to read in, and filling the entire 1,048,576-token window once costs $0.42. Switch the model field only — no other part of the request changes.

curl https://tokenra.io/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
            
curl https://tokenra.io/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"kimi-k3.1-pro","messages":[{"role":"user","content":"Summarize the supplied document and label each claim with its source line."}]}'
            

              Swap "model":"kimi-k3.1-pro" for
              "model":"kimi-k3.1-pro-0.2x" to bill the same request at 2折
              ($0.40 input and $2.00 output per 1M tokens). Nothing else in the request
              body changes.
            
            

              Both identifiers are enabled in the console today. If your account has not
              been provisioned for one of them, the call returns
              model_not_found before billing — that tells you about access, not
              about your payload.