minimax
MiniMax: MiniMax M3
minimax/minimax-m3
Our take
AI-assisted analysis based on the sources below. Suitability is an editorial assessment; test it on your own workload.
MiniMax M3 (minimax/minimax-m3) is a natively multimodal Mixture-of-Experts model from MiniMax, accepting text, image, and video inputs and producing text over a context window of about one million tokens. Vendor documentation describes a sparse-attention design aimed at long-context efficiency and three selectable reasoning modes. OpenRouter's catalog lists support for tools, structured outputs, and cached input pricing, plus aggregated benchmark entries in agentic and design categories. Evidence here is limited to the provider's model card and the catalog listing, so independent verification is advised before adoption.[1][2]
Strengths
- Native multimodal handling: text, image, and video inputs with text output, plus a context window of roughly one million tokens, per the vendor card and OpenRouter listing.[1][2]
- Vendor reports a sparse-attention architecture (MSA) intended to reduce compute and memory at very long contexts; these efficiency gains are vendor claims, not independently measured in this pack.[2]
- Flexible operation: three reasoning modes (enabled, adaptive, disabled) documented by the vendor, and API-level support for tools, structured outputs, and sampling controls listed in the catalog.[1][2]
Limitations
- All evidence comes from the vendor's own model card and OpenRouter's catalog; no independent evaluation, model card review, or safety assessment was supplied, so claims should be validated on real workloads.[1][2]
- Released under a custom 'minimax-community' license; the full terms were not included in this source pack and should be reviewed before deployment.[2]
- Aggregated benchmark entries in the catalog suggest mid-range results across agentic and front-end categories; treat these as indicative rather than definitive, and note the knowledge cutoff is unreported.[1]
Where to use it
- Worth evaluating for long-horizon agentic and coding workflows, which the vendor emphasizes and the large context window may support; confirm on your own tasks.[1][2]
- A candidate for multimodal pipelines that must read images or video and return text, given the natively multimodal training claimed by the vendor.[1][2]
- May suit API deployments needing tool calling or structured outputs with optional, adjustable reasoning; no self-hosting or hardware requirements were evidenced beyond vendor-suggested frameworks.[1][2]
Sources
- [1] OpenRouter catalog · MiniMax: MiniMax M3Sources checked: Nov 13, 2026
- [2] Model documentation · MiniMax: MiniMax M3Sources checked: Nov 13, 2026
About this model
Description below is the original text supplied by OpenRouter, shown in the source language.
MiniMax-M3 is a multimodal foundation model from MiniMax. It supports text, image, and video inputs with text output, a 1M-token context window, and is suited for long-horizon agentic work, coding,...
Specifications
- Model ID
- minimax/minimax-m3
- Canonical ID
- minimax/minimax-m3-20260531
- Creator
- minimax
- Added to OpenRouter
- 31.05.2026
- Context length
- 1,048,576
- Modality
- text+image+video->text
- Input modalities
- text, image, video
- Output modalities
- text
- Token pricing
- USD per 1M tokens: in $0.300 / out $1.2
- 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 | CC — below the B band | reasoning: optional · parameters: include_reasoning, reasoning |
| Coding | BB — next 30% of the cohort | — |
| Agents | AA — next 20% of the cohort | parameters: structured_outputs, tool_choice, tools |
| Writing | No measurement | output: text |
| Reasoning | No measurement | reasoning: optional · parameters: include_reasoning, reasoning |
| Document analysis | No measurement | — |
| Vision | No measurement | input: image |
| Image generation | No measurement | — |
| Audio | No measurement | — |
| Video | No measurement | input: video |
| Embeddings | No measurement | — |
Evaluations
| Source | Benchmark | Value | Grade | Data retrieved |
|---|---|---|---|---|
| Artificial Analysis | Intelligence index | 18.9 | CC — below the B band | 13.11.2026 |
| Artificial Analysis | Coding index | 58.6 | BB — next 30% of the cohort | 13.11.2026 |
| Artificial Analysis | Agentic index | 29.5 | AA — next 20% of the cohort | 13.11.2026 |
| Design Arena | agents · agenticgamedevwin rate 44.9% · rank 18 | 1,114Elo | CC — below the B band | 13.11.2026 |
| Design Arena | agents · androidnativewin rate 45.1% · rank 25 | 1,156Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | agents · blender3dwin rate 42.5% · rank 15 | 1,090Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | agents · codetovideowin rate 57% · rank 6 | 1,215Elo | AA — next 20% of the cohort | 13.11.2026 |
| Design Arena | agents · fullstackwin rate 47.4% · rank 19 | 1,179Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | agents · htmlslideswin rate 46.7% · rank 14 | 1,077Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | agents · mobileappswin rate 46.5% · rank 27 | 1,171Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | agents · pdfwin rate 61.6% · rank 7 | 1,261Elo | AA — next 20% of the cohort | 13.11.2026 |
| Design Arena | agents · python-pptxslideswin rate 47.5% · rank 16 | 1,166Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | agents · spreadsheetwin rate 51.2% · rank 17 | 1,155Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | agents · webappswin rate 47.7% · rank 22 | 1,192Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · 3dwin rate 50.1% · rank 52 | 1,193Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · asciiartwin rate 45.5% · rank 43 | 1,149Elo | CC — below the B band | 13.11.2026 |
| Design Arena | models · codecategorieswin rate 50.5% · rank 47 | 1,237Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · datavizwin rate 51% · rank 44 | 1,231Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · gamedevwin rate 45.4% · rank 55 | 1,202Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · svgwin rate 44% · rank 48 | 1,148Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · uicomponentwin rate 50.2% · rank 47 | 1,236Elo | BB — next 30% of the cohort | 13.11.2026 |
| Design Arena | models · websitewin rate 51.6% · rank 45 | 1,249Elo | 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) |
|---|---|---|---|---|---|---|
| DeepInfrafp8 | 524,288 | $0.224 | $0.880 | — | — | 85% |
| CoreWeavefp4 | 262,144 | $0.230 | $0.960 | — | — | 100% |
| GMICloudfp8 | 1,048,576 | $0.240 | $0.960 | — | — | 100% |
| DeepInfrafp8 | 524,288 | $0.280 | $1.1 | — | — | 100% |
| Venicefp8 | 524,288 | $0.300 | $1.2 | — | — | 100% |
| Parasailfp8 | 1,048,576 | $0.300 | $1.2 | — | — | 100% |
| Minimax | 524,288 | $0.300 | $1.2 | — | — | 100% |
| Parasailfp8 | 1,048,576 | $0.450 | $1.8 | — | — | 100% |
Provider data retrieved: 13.11.2026
Related articles
No published articles identify this model yet.