Short answer
For a new setup, don’t count on MiniMax M2.5 through Ollama. Ollama’s M2.5 library page says the model was retired on July 31, 2026, and that no models have been pushed. Its tags page does not show a runnable model tag. That means the official Ollama listing does not currently document a local M2.5 download or an Ollama Cloud tag. This is not proof that no community-made artifact exists, but the available official listing does not provide a verified command for one. (Ollama model listing; Ollama tags)
Ollama Cloud is hosted inference, not running the model on your own computer. A third-party setup guide may describe an earlier cloud integration, but it does not override the retirement status shown in Ollama’s listing. Don’t use an old M2.1 command as an M2.5 substitute: the captured Ollama command for M2.1 names a different model version and repository.
What this means for MiniMax API access
MiniMax’s current model notice says new users could no longer access M2.5 starting October 19, 2026. Existing pay-as-you-go API users can continue until November 18, 2026, when MiniMax says M2.5 API requests will stop working. For M Plan and Token Plan users, requests using M2.5 names have been routed to MiniMax-M3.1-Flash since October 19; the old names remain for backward compatibility. These are different access conditions, so check which account and billing plan you use before relying on an existing M2.5 integration. (MiniMax model list and retirement notice)
For a new MiniMax API integration, the current API documentation lists MiniMax-M3.1-Flash and MiniMax-M3 as supported language models. MiniMax documents OpenAI API format support and gives https://api.minimax.io/v1 as the OpenAI-compatible base URL. Use a currently supported model name rather than assuming MiniMax-M2.5 will work for a new account. (MiniMax API overview; OpenAI SDK guide)
Current listed API price example
MiniMax’s LLM pricing table lists MiniMax-M3.1-Flash at $0.15 per 1 million input tokens, $0.60 per 1 million output tokens, and $0.015 per 1 million prompt-cache read tokens. For a workload with 1 million uncached input tokens and 200,000 output tokens, the listed input/output charges total $0.27: $0.15 + (0.2 × $0.60). This illustration excludes cache-read charges and is a token-charge calculation, not a quote for a subscription plan or a forecast of actual usage. (MiniMax pricing overview)
Which route should you choose?
| Your situation | Practical choice | Important condition |
|---|---|---|
| You want to run M2.5 locally using Ollama | Don’t plan a deployment from the current official listing | It is marked retired, and the listing says no models have been pushed; no verified local M2.5 command or hardware requirements are documented there. (Ollama model listing) |
| You already call M2.5 through MiniMax pay-as-you-go API | Treat it as a migration window, not a new long-term deployment | MiniMax says existing users can access it only until November 18, 2026. (MiniMax model list and retirement notice) |
| You use MiniMax M Plan or Token Plan | Check that the M3.1-Flash behavior works for your application | MiniMax says M2.5-named requests under these plans have been routed to M3.1-Flash since October 19, 2026. (MiniMax model list and retirement notice) |
| You are starting a new production integration | Evaluate a currently listed API model, such as M3.1-Flash | Confirm the model, plan, limits, and current price in MiniMax’s documentation before deployment. (MiniMax API overview) |
Bottom line: The current evidence supports neither a fresh local Ollama setup for M2.5 nor a new MiniMax API deployment using M2.5. Existing API access is time-limited or routed to a successor depending on the plan. For new work, evaluate a currently supported model and test it against your own prompts, tool calls, latency, and budget.
Sources
MiniMax M2.5 Tags — Ollama Library
Model List and Retirement Notice — MiniMax API Docs