MiniMax-M2.5

MiniMax · MiniMax M

MiniMax's still-active M-series model for coding, tool use, and office-style agent workflows.

Part of MiniMax M family · Other versions: MiniMax M2.7 , MiniMax M2.5
Type
language
Context
205K tokens
Max Output
205K tokens
Status
current
Input
$0.3/1M tok
Output
$1.2/1M tok
API Access
Yes
License
proprietary
chinese reasoning coding agentic long-context
Released February 2026 · Updated May 16, 2026

Overview

Freshness note: Model capabilities, limits, and pricing can change quickly. This profile is a point-in-time snapshot last verified on May 16, 2026.

MiniMax-M2.5 is no longer the newest M-series model, but it remains an active MiniMax API tier. MiniMax’s official February 12, 2026 launch notes position it around real-world productivity: coding, tool use, search, and office-style workflows rather than only benchmark-first reasoning. For new MiniMax evaluations, start with MiniMax M2.7 and use this page for compatibility or generation-to-generation comparison.

MiniMax positions M2.5 as a research-to-production model, not just a benchmark release. That makes it relevant for real assistant and agent systems rather than only leaderboard tracking.

Capabilities

MiniMax-M2.5 is strongest in long-context reasoning, software engineering tasks, and structured answer generation for multi-step workflows. Official release notes emphasize gains in coding, search, tool use, and office-style evaluations relative to the previous M2.1 generation.

It is also well suited to retrieval-heavy assistants where the ability to reason across large context windows is operationally important.

Technical Details

MiniMax’s current docs keep M2.5 in the active text-model roster and the pay-as-you-go pricing page lists both MiniMax-M2.5 and MiniMax-M2.5-highspeed. Public docs still emphasize throughput and economic efficiency more than raw architecture disclosure, so this profile keeps the repo’s long-context snapshot while treating the model as a proprietary managed API product rather than an open-weight release.

Pricing & Access

MiniMax’s current pay-as-you-go pricing lists MiniMax-M2.5 at 0.30per1Minputtokensand0.30 per 1M input tokens and 1.20 per 1M output tokens, with prompt-caching read pricing at $0.03 per 1M tokens. MiniMax-M2.5-highspeed doubles the speed and doubles the token price.

Best Use Cases

Use MiniMax-M2.5 for coding copilots, document-heavy analytical assistants, and agent systems that need long-context synthesis with strong reasoning depth at a relatively low API price.

It is a strong fit when you want a lower-cost proprietary frontier-style model rather than an open-weight deployment.

Comparisons

  • Qwen3-Max (Alibaba): Qwen3-Max has deeper Alibaba ecosystem integration; MiniMax-M2.5 is attractive when you want MiniMax’s coding and productivity emphasis at a lower cost.
  • GLM-5 (Zhipu AI): GLM-5 offers competitive hosted pricing and high output limits; MiniMax-M2.5 is compelling when M-series productivity and coding benchmarks matter more.
  • DeepSeek-R1 / Reasoner (DeepSeek): DeepSeek remains a strong value option for pure reasoning economics; MiniMax-M2.5 is broader on coding, search, and office-style work.