minimax
MiniMax: MiniMax M2.5
minimax/minimax-m2.5
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
MiniMax M2.5 is a text-to-text large language model from MiniMax, listed on OpenRouter with a 200,000-token context window and mandatory reasoning. According to its model card, it was trained with reinforcement learning across hundreds of thousands of real-world environments, targeting coding, agentic tool use, search, and office productivity. The vendor reports state-of-the-art coding results and faster, more token-efficient agent runs than its predecessor M2.1, while OpenRouter's Design Arena shows mid-table placements. Independent verification of vendor claims is limited in this source pack.[1][2]
Strengths
- The vendor reports strong coding ability trained across more than ten programming languages and full-stack platforms, with an emergent architect-style planning step before code generation.[2]
- Agentic tool use and search are emphasized: the vendor reports fewer search rounds and better token efficiency than M2.1, and OpenRouter lists tool calling and structured output support.[1][2]
- Large 200,000-token context with mandatory reasoning; the vendor claims high serving throughput and very low operating cost for sustained agent workloads.[1][2]
Limitations
- Headline benchmark results (SWE-Bench, BrowseComp, an internal office-work evaluation) are self-reported by the vendor and not independently verified in this pack.[2]
- Text-only input and output; no multimodal modalities are listed in the catalog entry.[1]
- Independent measured signal is limited to mid-table Design Arena placements on OpenRouter; knowledge cutoff is unstated and the supplied card is truncated, so validate on your own workload.[1][2]
Where to use it
- Agentic coding workflows such as multi-language, full-stack development and code review, given vendor-reported strengths; pilot on your own repositories first.[1][2]
- Search and research agents plus tool-calling pipelines where throughput and cost efficiency matter.[2]
- Drafting office deliverables such as documents, spreadsheets, and presentations, suggested by the vendor's office-work training but not externally verified.[2]
Sources
- [1] OpenRouter catalog · MiniMax: MiniMax M2.5Sources checked: Nov 13, 2026
- [2] Model documentation · MiniMax: MiniMax M2.5Sources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
MiniMax-M2.5 is a SOTA large language model designed for real-world productivity. Trained in a diverse range of complex real-world digital working environments, M2.5 builds upon the coding expertise of M2.1...
Specifications
- Model ID
- minimax/minimax-m2.5
- Canonical ID
- minimax/minimax-m2.5-20260211
- Creator
- minimax
- Added to OpenRouter
- 12.02.2026
- Context length
- 200,000
- Modality
- text->text
- Input modalities
- text
- Output modalities
- text
- Token pricing
- USD per 1M tokens: in $0.270 / out $1.08
- Knowledge cutoff
- —
- Expiration date
- —
Tasks and capabilities
A grade appears only where the source supplies a matching measurement for that task, the model is active and the cohort holds at least 10 comparable models. The capability column lists facts the source declares — supported formats, reasoning and tool parameters — which are not measured quality.
| Task | Grade | Source-declared capability |
|---|---|---|
| General quality | No measurement | reasoning: mandatory · parameters: include_reasoning, reasoning |
| Coding | No measurement | — |
| Agents | No measurement | parameters: structured_outputs, tool_choice, tools |
| Writing | No measurement | output: text |
| Reasoning | No measurement | reasoning: mandatory · parameters: include_reasoning, reasoning |
| Document analysis | No measurement | — |
| Vision | No measurement | — |
| Image generation | No measurement | — |
| Audio | No measurement | — |
| Video | No measurement | — |
| Embeddings | No measurement | — |
Evaluations
| Source | Benchmark | Value | Grade | Data retrieved |
|---|---|---|---|---|
| Design Arena | models · 3dwin rate 57.6% · rank 67 | 1,153Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · codecategorieswin rate 56.8% · rank 64 | 1,198Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · datavizwin rate 51.2% · rank 71 | 1,170Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · gamedevwin rate 55.5% · rank 69 | 1,167Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · svgwin rate 54.5% · rank 52 | 1,141Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · uicomponentwin rate 53.4% · rank 76 | 1,163Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · websitewin rate 57.5% · rank 62 | 1,215Elo | BB — next 30% of the cohort | 13.11.2026 |
How grades work
Each letter grade compares a model only against other models scored by the same source and metric in the current snapshot. S means the top 10%, A the next 20%, B the next 30%, C the rest. Models without a published measurement are shown as not rated — a supported format alone never implies measured quality.
- S — S — top 10% of the cohort
- A — A — next 20% of the cohort
- B — B — next 30% of the cohort
- C — C — below the B band
Providers
Optional live provider statistics from OpenRouter. Missing values are unknown, not zero.
| Provider | Context length | Price in /1M | Price out /1M | Latency p50 (30m, s) | Throughput p50 (30m, tok/s) | Uptime (30m) |
|---|---|---|---|---|---|---|
| Venice | 198,000 | $0.270 | $0.950 | — | — | 83.33% |
| AtlasCloudfp8 | 196,608 | $0.295 | $1.2 | — | — | 100% |
| GMICloudfp8 | 196,608 | $0.300 | $1.2 | — | — | 100% |
| DigitalOcean | 65,536 | $0.300 | $1.2 | — | — | 100% |
Provider data retrieved: 13.11.2026
Related articles
No published articles identify this model yet.