Skip to main content

​
MiniMax M2.5

Pick this for long-context reasoning, multi-step planning, and tool calling — a 230B MoE reasoner whose explicit reasoning output surfaces in the playground.
Model ID: MiniMaxAI/MiniMax-M2.5

​
About

MiniMax M2.5 is a 230B MoE reasoning transformer that emits reasoning chunks before the final content for transparency. It is a text-only model — image inputs are rejected by the gateway.

​
Pricing

DirectionRate (USD)
Input$0.45 / 1M tokens
Output$1.80 / 1M tokens

​
Capabilities

  • chat
  • reasoning
  • tool-use

​
Specs

  • Context window: 196k tokens
  • License: Apache-2.0
  • Released: March 8, 2026
  • Chat and instruction following
  • Multi-step reasoning and planning
  • Tool use and function calling

​
Benchmarks

Benchmarks fill in during Alpha. Tracked as a follow-up to RD-562.

​
Example prompts

  • Walk through the trade-offs of three caching strategies for a read-heavy API and recommend one.
  • Given this Go stack trace, identify the most likely root cause and suggest 3 next debugging steps.

​
Quick start

curl https://api.corvex.ai/v1/chat/completions \
  -H "Authorization: Bearer $CORVEX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "MiniMaxAI/MiniMax-M2.5",
    "messages": [{"role": "user", "content": "Hello"}]
  }'
Open in the Corvex playground