Direct answer
“MiniMax-M2 5 parameters” is ambiguous: “M2 5” could mean MiniMax-M2.5, while “parameters” could mean a model’s learned weights or API generation settings. The query alone does not tell us which meaning or version you intend.
For MiniMax-M2—not M2.5—the model repository describes 230 billion total parameters, with 10 billion active parameters. It recommends three inference values: temperature=1.0, top_p=0.95, and top_k=40. Those are repository recommendations, not a verified five-setting production API schema or proof of API defaults. (MiniMax-M2 repository; README)
The supplied M2.5 material identifies MiniMax-M2.5 as a distinct version, but does not establish its parameter count. It also does not establish that “5 parameters” refers to five API settings. (M2.5 listing)
What “parameters” means here
In a model description, parameters are learned model weights. The 230B total / 10B active figure applies to MiniMax-M2 as described in its repository; it should not be carried over to M2.5.
In an API request, generation settings are fields that influence how output is generated. They are separate from the model’s learned parameter count. For MiniMax-M2, the repository recommends these three settings: (MiniMax-M2 README)
| Setting | M2 repository recommendation | Practical meaning |
|---|---|---|
temperature |
1.0 |
Sampling control that affects output variation. |
top_p |
0.95 |
Sampling control based on cumulative token probability. |
top_k |
40 |
Sampling control that limits selection to a set of candidate tokens. |
These are the repository’s recommended values—not documented universal defaults or a five-setting recipe. The supplied recommendation does not give a complete set of valid ranges or establish that every hosted provider exposes these fields in the same way.
Why this is not a verified five-setting API list
Vals AI reports that its M2.5 benchmarks used the “lightning” endpoint, temperature=1, top_p=0.9, and a maximum output of 196,608 tokens. That is a third-party report of a benchmark setup—not an official API schema, not evidence of M2 defaults, and not a list of five API fields. (Vals AI’s M2.5 report)
If you need five fields for a request, use the schema for the specific model, API route, and provider you plan to deploy with. Don’t add a plausible field such as max_tokens or stream and call it a verified M2 setting without documentation for that route.
Production selection: what to check
The M2 repository positions MiniMax-M2 for coding and agentic workflows. That positioning can inform an evaluation, but it does not guarantee results on your codebase, agent framework, or production traffic. Test representative prompts and tools against your own quality, latency, and failure criteria before choosing it. (MiniMax-M2 repository)
MiniMax’s official model list currently names M3.1-Flash and M3 in its available LLM list. Its retirement notice specifically covers M2.5 and M2.7, with retirement scheduled for November 18, 2026; that notice does not establish M2’s status. The retrieved model-list information therefore does not confirm whether MiniMax-M2 itself is currently callable through the hosted API. Nor does it verify that M2’s repository-recommended temperature, top_p, and top_k settings are accepted on a hosted M2 route. Confirm the exact model identifier, availability, supported request fields, limits, and pricing for your intended API channel before shipping. (MiniMax API model list)
Bottom line: If you mean MiniMax-M2’s model size, its repository describes 230B total / 10B active parameters. If you mean M2 generation settings, the repository recommends three values—not five. If you mean MiniMax-M2.5, the available material here does not establish its parameter count. Verify settings and availability against the exact production model and API route.