Kimi K3.1 Pro Is Live, a Third Cheaper Than K3

moonshot/kimi-k3.1-pro is live on BlockRun as of today, October 25, on Base and on Solana. It is Moonshot's new flagship. Moonshot built it by post-training Kimi K3 for long-horizon coding and agent work, and it keeps K3's 1M-token context window.

It reasons, it reads images, and it calls tools. Send it through the same OpenAI-compatible endpoint as every other model. You don't need a Moonshot account or a key of your own.

What it is for

K3.1 Pro is an open-weight model tuned for work that runs a long time:

• Coding agents. Plan a change, edit files, run the tests and read the output across many turns. With a 1M-token window, a large codebase fits in the prompt.
• Tool-heavy workflows. It calls tools when you leave the choice to it, and calls the one you name when you force tool_choice.
• Reading screenshots and documents. Image input works next to text in the same message.

It returns its reasoning in reasoning_content, so you can log it or show it without parsing the answer.

What it costs

$2 in, $10 out per 1M tokens, $0.20 for cached input. That is Moonshot's own rate. We pass LLM tokens through at cost, so you pay what Moonshot charges.

That is a third less than Kimi K3 on both input and output. If you already run K3, moving to K3.1 Pro is a one-line change to the model id, and the bill drops by a third on every token.

Reasoning tokens bill as output, at the $10 rate. Give reasoning-heavy calls room in max_tokens.

What we checked before listing it

We don't list a model on a 200 from a models endpoint. Before K3.1 Pro went into the catalog, we sent it real requests:

• a plain completion, which came back with its reasoning in reasoning_content
• a streamed completion
• an automatic tool call, where the model chose to call the tool
• a forced tool call, where we named the tool in tool_choice
• an image, which it described correctly

All five came back right.

Kimi K3

K3.1 Pro replaces K3 as the Kimi we feature. moonshot/kimi-k3 stays routable at its own price, so nothing you have already shipped breaks. It is no longer on the public model list.

Calling it

The chat endpoint is OpenAI-compatible, and the SDK handles the payment:

from blockrun_llm import LLMClient

client = LLMClient()  # wallet from BLOCKRUN_WALLET_KEY or ~/.blockrun/.session
print(client.chat("moonshot/kimi-k3.1-pro", "Read this repo's failing test and propose the fix."))

On Solana, make the same call with the Solana client, through the sol.blockrun.ai gateway:

from blockrun_llm import SolanaLLMClient

client = SolanaLLMClient()  # wallet from SOLANA_WALLET_KEY; pays USDC on Solana
print(client.chat("moonshot/kimi-k3.1-pro", "Summarize what changed in this diff and why."))

Or with curl:

curl https://blockrun.ai/api/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-PAYMENT: $PAYMENT_HEADER" \
  -d '{
    "model": "moonshot/kimi-k3.1-pro",
    "messages": [{"role": "user", "content": "Write a migration plan for this schema change."}]
  }'

The first response is a 402 that carries the exact price. Sign it and retry, and the answer comes back. Teams can use an API key and pay by card or invoice instead. An autonomous agent can pay per call from its own wallet in USDC on Base or Solana over x402, with no account at all.

Try it

The full catalog, with live prices, is at blockrun.ai/models. The docs cover each model's request rules. If your agent runs on K3 today, change one line and run your next task on K3.1 Pro.