MiniMax M3 is an open-weight Mixture-of-Experts model for agentic coding and computer-use agents, built around MiniMax Sparse Attention so a very large context window stays fast to decode. It succeeds MiniMax's prior flagship as the company's top model and ships with published weights rather than a closed API only.
Released June 1, 2026, MiniMax M3 is an open-weight model built around MiniMax Sparse Attention, which keeps decoding fast even at its full 1M-token context window. It accepts text, image, and video input, though not audio, and scores 70.06% on OSWorld-Verified for autonomous computer-use tasks.
Where it sits
- $0.525/M$ per 1M tokensBlended price (3:1)Lower is better#13 / 59peer median $2.00/Mvendor price, checked by HokAI
- --tokens/sOutput speedHigher is better-- / 33peer median 90 tok/scited: Artificial Analysis
- --% solvedSWE-bench VerifiedHigher is better-- / 26peer median 78.3%per source, see benchmark scores
- --% correctGPQA DiamondHigher is better-- / 41peer median 86.9%per source, see benchmark scores
Cheaper than 76% of the 59 GA models with a published price, and one of 62 whose vendor states it does not train on customer data.
Ranks are against GA models on HokAI that publish the same figure; ties share a rank.
Provider: MiniMax · Family: MiniMax M3
Context window: 1,048,576 tokens · Max output: 512,000
Input modalities: text, image, video, tool-calls · Output: text, tool-calls
About MiniMax M3
MiniMax M3 is the latest large language model from MiniMax, a Shanghai-based AI lab listed on the Hong Kong Stock Exchange (0100.HK) since January 2026. MiniMax released M3 on June 1, 2026, as a Mixture-of-Experts model with roughly 229.9 billion total parameters and 9.8 billion active parameters per token across 256 fine-grained experts. The headline architectural change is MiniMax Sparse Attention (MSA), a new sparse attention scheme that replaces the full-attention backbone used in the prior M2 generation. M3 sits at the top of MiniMax's M-series lineup, succeeding MiniMax-M2.7 (released March 18, 2026, open-weighted in April 2026), and was designed to solve three problems at once: make 1M-token context windows computationally practical, add native multimodal input from the start of pretraining rather than as an afterthought, and push agentic coding performance past frontier US and Chinese rivals at a fraction of the price. On MiniMax's own June 2026 release benchmarks, M3 scores 59.0% on SWE-Bench Pro, which the company says surpasses both GPT-5.5 and Gemini 3.1 Pro on the same test. It also scores 66.0% on Terminal-Bench 2.1, 74.2% on MCP Atlas (a tool-use benchmark), 70.06% on OSWorld-Verified (computer-use), 34.8% on SWE-fficiency, and 28.8% on KernelBench Hard. On multimodal evaluations, M3 reaches 84.8% on Video-MME, 81.4% on VideoMMMU, 75.1% on MMMU Pro, and 80.8% on OmniDocBench. On math olympiad benchmarks it solved 25 of 42 problems on IMO 2025 and 17 of 42 on USAMO 2026. A widely cited BrowseComp score of 83.5 is reported to beat Claude Opus 4.7's 79.3 on the same autonomous-browsing benchmark. Some third-party aggregators list a GPQA Diamond score near 92.9%, but that figure does not appear in MiniMax's own benchmark table and should be treated as unverified until MiniMax publishes its technical report. M3's headline feature is a 1,048,576-token (1M) context window with a 512,000-token maximum output, both roughly 5x the 204,800-token context of M2.7. The MSA architecture is designed specifically to make that window usable: MiniMax reports a 9.7x speedup in prefill and a 15.6x speedup in decode at 1M tokens compared to M2's full-attention design, cutting per-token compute at full context to roughly a twentieth of the previous generation. MSA keeps a Grouped-Query Attention backbone but adds block-level selection over real, uncompressed key-value pairs, which MiniMax contrasts with DeepSeek's Multi-head Latent Attention approach of compressing keys and values into a low-dimensional latent space. Independent time-to-first-token figures from Artificial Analysis put M3 at around 2.59 seconds on the MiniMax-hosted endpoint, with output speed around 54.8 tokens per second, which is workable for agentic loops but not the fastest option for latency-sensitive chat. M3 is natively multimodal for input, accepting text, images, and video, with text as the only confirmed output modality. It supports function calling, structured outputs, and an optional 'thinking' reasoning mode that can be toggled per request at no extra price premium. MiniMax also demonstrated computer-use: M3 can operate a desktop environment directly, scoring 70.06% on OSWorld-Verified. Despite MiniMax's broader audio product line (Speech 2.8, Music 2.5+), M3 itself does not support audio input or output, so voice workflows require pairing it with a separate MiniMax Audio endpoint or third-party ASR/TTS. MiniMax trained these modalities in together from the start of pretraining using interleaved text, image, and video data, which the company says scales better than bolting on synthetic multimodal data after the fact. M3 is available directly through MiniMax's own API platform (platform.minimax.io) and MiniMax Code, plus third-party inference providers including OpenRouter, Fireworks AI (with Day-0 support), and Together AI. As an open-weight model, M3's weights and technical report were published on Hugging Face (MiniMaxAI/MiniMax-M3) and GitHub (MiniMax-AI/MiniMax-M3) within about ten days of the June 1, 2026 API launch, under a MiniMax Community License similar in spirit to the modified-MIT terms used for M2.7. At 229.9B total parameters, self-hosting the full BF16 weights requires roughly 480GB or more of GPU memory, realistically a multi-node cluster of 8x H100/H200-class GPUs, though FP8 quantization can reduce that footprint. No AWS Bedrock, Google Vertex, or Azure listing was found for M3 as of mid-June 2026; MiniMax's infrastructure spans North America, Europe, and Asia-Pacific data regions. MiniMax has not published a system card, responsible scaling policy, or named third-party red-teaming partners for M3, consistent with the company's pattern for the M-series generally. Safety behavior is governed by MiniMax's general API terms of service rather than a dedicated model safety report, and the company states that customer API inputs are not stored or used for training by default unless a customer opts in. MiniMax has not published a SOC 2 Type II attestation, an ISO 27001 certificate, or HIPAA-eligible status for its platform, which limits M3's appeal for regulated US and EU enterprise workloads compared with Anthropic, OpenAI, or Google models that publish detailed model and system cards. MiniMax has not disclosed a specific training data cutoff date or detailed dataset composition for M3, continuing the pattern from its prior generation. The model was trained with native multimodal interleaved text, image, and video data rather than text-first pretraining followed by multimodal fine-tuning. As a Shanghai-headquartered, Hong Kong-listed company, MiniMax operates under China's Generative AI Measures from the Cyberspace Administration of China alongside any GDPR obligations for EU-facing products. Given MiniMax's cadence of roughly one major M-series release per quarter (M2.5 in February, M2.7 in March, M3 in June 2026), a further update is plausible later in 2026. MiniMax has not announced an M3.1 or successor as of mid-June 2026, and M2.7 remains available for users who do not need the larger context window or multimodal input.
Pricing
MiniMax M3's standard API pricing is $0.30 per 1M input tokens, $1.20 per 1M output tokens, and $0.06 per 1M cached input tokens for prompts up to 512K input, confirmed on MiniMax's own platform docs and OpenRouter as of August 2026. MiniMax's official pricing page states this is a permanent 50% off discount (list rate $0.60/$2.40/$0.12), not a temporary launch promotion. The long-context tier for inputs above 512K is exactly double the standard tier: $0.60 per 1M input, $2.40 per 1M output, $0.12 per 1M cached input. Monthly Token Plans (Plus $20, Max $50, Ultra $120) pool usage across text, image, speech, and music.
What a real job costs
| Job | Input | Output | Total |
|---|---|---|---|
| Summarise a 20-page PDF | $0.0090 | $0.0012 | $0.010 |
| Support reply | $0.0006 | $0.0004 | $0.0010 |
| One coding agent run | $0.060 | $0.024 | $0.084 |
Budgets: 20-page PDF = 30k in / 1k out · Support reply = 2k in / 300 out · Coding agent run = 200k in / 20k out. Computed from the vendor's per-token prices at render time; cached-input discounts are not applied.
Key Features
- MiniMax Sparse Attention (MSA): Replaces the full-attention backbone of M2 with block-level selection over real, uncompressed key-value pairs, unlike DeepSeek's approach of compressing them into a latent space.
- 1,048,576-Token Context Window: 5x the 204,800-token window of M2.7, with a 9.7x faster prefill and 15.6x faster decode at full context thanks to MSA.
- Native Multimodal Input + Computer Use: Accepts text, image, and video natively (trained in from pretraining), and can operate a desktop environment directly for computer-use tasks.
- Optional Thinking Mode: Per-request toggle for visible multi-step reasoning at the same per-token price, useful for coding and agentic planning tasks.
- Open Weights on Hugging Face: Released as MiniMaxAI/MiniMax-M3 under a MiniMax Community License, allowing self-hosting and fine-tuning roughly 10 days after the API launch.
Pros
- Among the cheapest frontier-tier agentic coding models, at roughly a twentieth of typical frontier pricing per token.
- 1M-token context window that stays fast at full length thanks to MiniMax Sparse Attention, making long-document agent workflows practical.
- Open weights (MiniMaxAI/MiniMax-M3 on Hugging Face) allow self-hosting and fine-tuning, unlike closed competitors that require using their hosted API only.
Cons
- No native audio input or output modality.
- No SOC 2 or HIPAA-eligible certification published, limiting appeal for regulated enterprise workloads.
- Self-hosting the 229.9B-parameter model requires a multi-GPU cluster (roughly 480GB+ VRAM at BF16).
Benchmarks
- Imo 2025: 59.5 vendor-reported · 01 Jun 2026
- MMMU-Pro: 75.1% vendor-reported · 01 Jun 2026 — College-level questions that need reading images and diagrams, % correct.
- MCP Atlas: 74.2% vendor-reported · 01 Jun 2026 — Tasks solved by calling external tools over the Model Context Protocol, % solved.
- Video Mme: 84.8 vendor-reported · 01 Jun 2026
- Usamo 2026: 40.5 vendor-reported · 01 Jun 2026
- Video Mmmu: 81.4 vendor-reported · 01 Jun 2026
- Browse Comp: 83.5 vendor-reported · 01 Jun 2026
- Omnidocbench: 80.8 vendor-reported · 01 Jun 2026
- SWE-bench Pro: 59% vendor-reported · 01 Jun 2026 — Harder, longer real-repository coding tasks, % solved.
- Swe Fficiency: 34.8 vendor-reported · 01 Jun 2026
- Kernelbench Hard: 28.8 vendor-reported · 01 Jun 2026
- OSWorld Verified: 70.1% vendor-reported · 01 Jun 2026 — Tasks completed by operating a real desktop, % solved.
- Terminal-Bench 2.1: 66% vendor-reported · 01 Jun 2026 — Multi-step tasks completed in a real command line, % solved.
A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.
Frequently Asked Questions
How much does MiniMax M3 cost?
Standard-tier input costs $0.30 per 1M tokens, output costs $1.20 per 1M, and cached input costs $0.06 per 1M for prompts up to 512K, confirmed on OpenRouter and MiniMax's own platform docs. MiniMax describes this as a permanent 50% discount off list pricing rather than a launch promotion, and anything past the 512K threshold bills at exactly double those rates. Teams who prefer flat billing can subscribe to a Token Plan instead, three tiers from $20 to $120 a month pooling usage across MiniMax's text, image, speech, and music models.
Is there a free way to try MiniMax M3?
MiniMax has not published a free tier for M3's API. The lowest-cost path is the $20/month Plus Token Plan, which pools a shared allotment across several MiniMax product lines rather than gating M3 specifically. Third-party hosts such as OpenRouter sometimes run short promotional credits, but none are confirmed as ongoing free access to M3.
What are the best alternatives to MiniMax M3?
MiniMax positions M3 against GPT-5.5 and Gemini 3.1 Pro on agentic coding and autonomous-browsing benchmarks, and against Claude Opus 4.7 specifically on BrowseComp. Teams who need a comparable 1M-token context without self-hosting typically weigh those three closed frontier models; teams who want another open-weight option would look at MiniMax's own prior-generation model, or wait for a same-generation release from DeepSeek, Qwen, or Kimi.
How does MiniMax M3 compare to GPT-5.5 and Gemini 3.1 Pro?
MiniMax's release blog reports M3 beating both on SWE-Bench Pro at 59.0%, though it did not publish either rival's exact score alongside that number. All of these figures are vendor-reported and have not yet been reproduced on an independent leaderboard, so the comparison should be read as MiniMax's own framing until a third party verifies it. A circulating ~92.9% GPQA Diamond figure for M3 does not appear in MiniMax's own release materials and should be treated as unconfirmed.
How do I get started with MiniMax M3?
M3 is reachable through MiniMax's own API at platform.minimax.io, or through third-party hosts including OpenRouter, Fireworks AI, and Together AI, all using a standard API-key setup. Teams that want to self-host instead can pull the published weights from Hugging Face (MiniMaxAI/MiniMax-M3) under the MiniMax Community License, though that requires the substantial multi-GPU hardware detailed above.
Top Alternatives
- MiniMax M2.7: Pick M3 for its far larger context window and native image and video input, both missing from M2.7; pick M2.7 only if its context ceiling already covers the job, since the two price identically per token.
- DeepSeek-V4-Pro: Pick M3 for roughly a fifth of DeepSeek-V4-Pro's per-token price; pick DeepSeek-V4-Pro if you want a fully permissive MIT license instead of MiniMax's own community license terms.
- GPT-5.5: MiniMax says M3 beats GPT-5.5 on SWE-Bench Pro at a fraction of its per-token price; pick GPT-5.5 for native audio input, which M3 lacks.
- Claude Opus 4.7: MiniMax reports M3 outscoring Opus 4.7 on its BrowseComp autonomous-browsing test; pick Opus 4.7 instead for a published system card and enterprise certifications, which M3 lacks.