AI model reference · 2026
MiniMax M3
MiniMax-M3 is a multimodal foundation model from MiniMax.
- Context window
- 1,048,576 tokens
- Maximum output
- 512,000 tokens
- Base provider API input
- $0.30 / 1M tokens
- Base provider API rate
- Published evidence
- 4 benchmarks, 2 Arena results
Overview
About MiniMax M3
A concise catalog overview. Technical limits and published evaluation evidence are listed separately below.
MiniMax-M3 is a multimodal foundation model from MiniMax.
Published evaluations
Benchmark results
Only results mapped to this exact model and backed by a named source appear here. Different evaluation protocols are not treated as interchangeable.
| Benchmark | Score | Evaluation details | Source |
|---|---|---|---|
| OSWorld-VerifiedA verified computer-use benchmark in which multimodal agents operate desktop applications and are graded from the resulting environment state. | 75.2% | Mean task rewardVersion: OSWorld-Verified v1 · Split: Up to 361 tasks, excluding eight Google Drive tasks where required · Harness: Official OSWorld unified environment · Attempts: One run · Tools: GUI interactionProtocol note: Only 100-step general-model rows are included in this normalized snapshot. Individual evaluated-task counts can differ when a task could not run.; Run date: 2026-06-07; Evaluated tasks: 358Methodology | OSWorldBenchmark owner |
| GPQA DiamondThe highest-quality subset of Graduate-Level Google-Proof Q&A, designed to test expert-level scientific reasoning in biology, physics, and chemistry. | 92.9% | AccuracyVersion: GPQA Diamond 198 · Split: 198 Diamond questions · Harness: Artificial Analysis independent evaluation · Tools: Not disclosed on the score pageMethodology | Artificial AnalysisIndependently reproduced |
| Humanity's Last ExamA 2,500-question expert-level benchmark spanning dozens of academic fields. Tool-assisted and no-tools results are separate protocols and must not be merged. | 39.0% | AccuracyVersion: May 2025 text-only revision · Split: 2,158 text-only questions from the 2,500-question May 2025 revision · Attempts: pass@1 · Tools: No browser or retrieval tools · Evaluator: LLM equality checker with numerical toleranceMethodology | Artificial AnalysisIndependently reproduced |
| LiveBenchA contamination-resistant benchmark refreshed on a fixed release cadence. Scores from different LiveBench releases must never be compared as the same protocol. | 67.3% | Mean of category averagesVersion: LiveBench 2026-06-25 · Split: 2026-06-25 release, 23 tasks across seven categoriesProtocol note: Overall is the mean of category averages. This protocol is not comparable with the 2026-01-08 release.Methodology | LiveBenchBenchmark owner |
Human preference
Arena results
Arena scores come from blind human preference votes. They are reported separately from task benchmarks and are not used as substitutes for missing benchmark results.
| Category | Score | Rank | Votes | Status |
|---|---|---|---|---|
| Arena TextBlind human preference score for text responses. | 1,444±5 | #72 | 37,312 | Published |
| Arena WebDevBlind human preference score for code and web development. | 1,490±7 | #34 | 10,053 | Published |
Technical reference
Specifications and API facts
Technical limits and reference provider prices come from the current catalog unless a reviewed external source is listed. Missing values stay marked as unavailable.
- Provider
- MiniMax
- Release date
- Not available in reviewed sources
- Catalog added
- May 31, 2026
- Context window
- 1,048,576 tokens
- Maximum output
- 512,000 tokens
- Knowledge cutoff
- Not available in reviewed sources
- License
- Not available in reviewed sources
- Input types
- Text, Image, Video
- Output types
- Text
- Reasoning
- Supported
- Provider endpoint tool parameters
- Accepts tool parameters
- Base provider API input
- $0.30 / 1M tokens
- Base provider cached input
- $0.06 / 1M tokens
- Base provider API output
- $1.20 / 1M tokens
These are reference provider API rates, not Writingmate checkout charges. Access in Writingmate follows the allowances of your Writingmate plan.
Provenance
Sources and update status
Source links are attached to the facts and evaluations they support. Catalog-only values are not presented as independently verified claims.
- Arena TextArena AI · Retrieved Aug 19, 2026
- Arena WebDevArena AI · Retrieved Aug 19, 2026
- GPQA Diamond Benchmark LeaderboardArtificial Analysis · Retrieved Aug 9, 2026
- Humanity's Last Exam Benchmark LeaderboardArtificial Analysis · Retrieved Aug 9, 2026
- LiveBench 2026-06-25 LeaderboardLiveBench · Retrieved Aug 9, 2026
- OSWorld-Verified ResultsOSWorld · Retrieved Aug 9, 2026
- OpenRouter model catalogOpenRouter
Evidence last updated Aug 19, 2026.
Catalog record updated May 31, 2026.
Catalog-added dates describe when a model entered the catalog, not necessarily its public release date.
Keep exploring
Related models and comparisons
Open another model’s detail page or compare it directly with MiniMax M3. Suggestions come from the current catalog.
GLM 5.3
Z.AI
GLM-5.3 is a large-scale reasoning model from Z.ai, built for complex software engineering and long-horizon agent tasks.
Qwen3.8 27B
Qwen
Qwen3.8 27B is an open-weight dense vision-language model from Qwen.
Seedream 5.0 Lite
ByteDance
Seedream 5.0 Lite is an image generation model from ByteDance Seed.
Gemini 3.7 Flash
Gemini 3.7 Flash is a multimodal model from Google for fast agentic workflows, coding, and complex multi-step reasoning.
Seedream 5.0 Pro
ByteDance
Seedream 5.0 Pro is an image generation and editing model from ByteDance Seed.
Seedance 2.0 Mini
ByteDance
Seedance 2.0 Mini is a video generation model from ByteDance.
Try MiniMax M3 in Writingmate
Use this model, then compare its response with other AI models in the same workspace.