Overview
MiniMax-M3 is MiniMax's native-multimodal flagship, released on June 1, 2026. MiniMax's own API release notes date it "Jun. 1, 2026," describing a model for "agentic reasoning, tool use, coding, multimodal chat input, and long-context tasks." The Hugging Face model card summarizes it in one line: "MiniMax-M3 is a native multimodal model with 1M context. It has ~428B parameters and ~23B activated parameters."
The architectural bet is sparse attention. M3 is the first MiniMax flagship built on MiniMax Sparse Attention (MSA), which the model card calls "a high-performance sparse attention operator designed for million-token contexts." In the launch blog, MiniMax claims the payoff plainly: "at a context length of 1 million, M3's per-token compute is just 1/20 that of the previous-generation model," with a speed up of "more than 9× in the prefilling stage and more than 15x in the decoding stage." The technical report, arXiv:2606.13392, landed on June 11, 2026 — ten days after the model.
The second bet is native multimodality. Rather than bolting a vision encoder onto a finished text model, MiniMax states that "M3 is a model that has undergone mixed-modality training from Step 0. This native multimodal approach allows the semantic spaces of different modalities to merge more naturally and deeply." Hugging Face classifies the repository as image-text-to-text; it accepts text, image, and video, and emits text.
M3 sits at the end of a fast release cadence documented in MiniMax's release notes: MiniMax-Text-01 (Jan 15, 2025) → M2 (Oct 27, 2025) → M2.1 (Dec 22, 2025) → M2.5 (Feb 2026) → M2.7 (Mar 18, 2026) → M3 (Jun 1, 2026).
A note on the license
MiniMax-M3 is not MIT licensed, and the "open source" framing does not survive contact with the license text.
MiniMax's launch blog promised to "open-source the corresponding model weights." What shipped is the MINIMAX COMMUNITY LICENSE. Hugging Face tags the repository license: other with license_name: minimax-community. The base grant covers use "for non-commercial purposes"; any Commercial Use triggers two conditions. The license requires that "you shall prominently display 'Built with MiniMax M3' on a related website, user interface, blogpost, about page or product documentation," and that organizations whose products "generate more than 20 million US dollars (or equivalent in other currencies) in yearly revenue" must "obtain a separate, prior written authorization from MiniMax by contacting api@minimax.io" — below that threshold a one-time notice to the same address suffices. It further prohibits military use, exploitation of minors, harmful misinformation, and hate speech. There are no geographic restrictions.
Under the definition used on this site, that is open weights under a restricted license — not open source.
The regression did not begin with M3. MiniMax-M2 was released under MIT in October 2025; today its Hugging Face card carries license: other with license_name: modified-mit, as does M2.5. Decrypt reported on April 13, 2026 that with M2.7 "commercial use now requires written authorization from MiniMax," quoting MiniMax developer-relations head Ryan Lee on degraded third-party hosting: "A fully permissive license meant we had no way to push back on any of that." Artificial Analysis noted the same in its June 8, 2026 article: "When MiniMax released the weights for M2.7, it was under a commercially restricted license." M3 continues that policy under a new name.
Capabilities
- Million-token context via MSA:
max_position_embeddingsin config.json is exactly 1,048,576. MSA is a blockwise sparse attention mechanism over a Grouped Query Attention backbone, not a compressed-latent scheme. - Native multimodal input: Text, image, and video in a single multimodal model trained mixed-modality from step zero, with a CLIP-style vision tower.
- Sparse MoE efficiency: A 428B-parameter mixture of experts activating ~23B per token — 128 routed experts, top-4 routing, plus one shared expert.
- Three-way reasoning control: A
thinkingparameter acceptingenabled,adaptive, ordisabled, letting one deployment span deep chain-of-thought and low-latency turns. - Agentic tool use: MiniMax reports 74.2% on MCP Atlas, a benchmark of Model Context Protocol tool orchestration, and 66.0% on Terminal-Bench 2.1.
- Long-horizon coding: MiniMax positions M3 as reaching "frontier-level performance across long-horizon agentic benchmarks, excelling in both coding and cowork."
- Broad serving support: SGLang, vLLM, Transformers (
minimax_m3_vl), KTransformers, and unsloth all ship documented recipes.
Technical Specifications
- Hugging Face repository:
MiniMaxAI/MiniMax-M3 - Total parameters: ~428B per the model card; 427,040,140,160 exactly per Hugging Face's safetensors metadata
- Active parameters: ~23B per token (model card); NVIDIA's deployment blog states 22B
- Experts: 128 routed, top-4 activated per token, plus 1 shared expert (config.json)
- Layers: 60 · Hidden size: 6,144 · Attention heads: 64 · KV heads: 4 (GQA)
- Vocabulary: 200,064 tokens · Weights dtype: bfloat16
- Vision tower: CLIP-style encoder, 32 layers, hidden size 1,280
- Architecture class:
MiniMaxM3SparseForConditionalGeneration(model_type: minimax_m3_vl) - Attention: MiniMax Sparse Attention (MSA), arXiv:2606.13392
- Context window: 1,048,576 tokens
- License: MiniMax Community License (
license: other) - Recommended sampling:
temperature=1.0,top_p=0.95 - Reasoning control:
thinking=enabled|adaptive|disabled
MiniMax does not publish a knowledge cutoff or a training-token count for M3, so this page does not state one.
Use Cases
- Whole-repository code work: A 1M-token context window admits large codebases in one pass, which is the workload MSA was built to make affordable.
- Long-horizon agents: Agentic workflows that accumulate long tool-call transcripts benefit most from the claimed 1/20 per-token compute at 1M context.
- Document and video understanding: Native video and image input, evaluated by MiniMax on Video-MME at 512 frames, suits multimodal archive and media analysis.
- Tool-calling backends: The
thinking: adaptivemode plus MCP-oriented evaluation targets function calling services that mix trivial and hard requests. - Cost-sensitive inference at scale: $0.30 / $1.20 per million tokens is materially below most frontier-adjacent hosted APIs.
- Self-hosting under compliance review: Weights are downloadable, but the $20M revenue authorization gate makes this a legal question before it is an engineering one.
Performance / Benchmarks
Vendor-reported. Every figure below appears as text in MiniMax's own launch blog. The model card's benchmark comparison is published only as an image (figures/benchmark.jpeg) and the repository's Hugging Face model-index is null. The repo does carry a .eval_results/minimax-m3.yaml, added in June 2026, but it states on its face that it was "Extracted from the model card benchmark graph" — a transcription of MiniMax's chart, not an audit of it.
| Benchmark | MiniMax-M3 (vendor-reported) |
|---|---|
| SWE-Bench Pro | 59.0% |
| MCP Atlas | 74.2% |
| OSWorld-Verified | 70.06% |
| Terminal-Bench 2.1 | 66.0% |
| SWE-fficiency | 34.8% |
| KernelBench Hard | 28.8% |
| Video-MME (512 frames) | 84.6 |
SWE-Bench Pro, MCP Atlas, Terminal-Bench 2.1, SWE-fficiency, and KernelBench Hard appear in the blog's headline list; OSWorld-Verified 70.06% and the 84.6 Video-MME result at 512 frames appear only in its evaluation-methodology notes.
MiniMax's blog names SWE-Bench Verified, Claw-Eval, MMMU Pro, VideoMMMU, BrowseComp, PaperBench, Apex-Agents, IMO 2025, and USAMO 2026 without publishing M3 scores in text for them. The repo's .eval_results/minimax-m3.yaml transcribes four of them off the benchmark JPEG — SWE-Bench Verified 80.5, MMMU-Pro 78.1, Claw-Eval 74.5, Apex-Agents 27.7 — along with a Video-MME (w/ sub) score of 85.4. As chart readings of MiniMax's own image, these carry no more weight than the vendor numbers above.
Third-party placements
Artificial Analysis reports 44 on the Artificial Analysis Intelligence Index v4.1, an aggregate of nine evaluations, with output throughput of 94.7 tokens per second. On the same v4.1 scale, the Intelligence Index comparison chart on M3's Artificial Analysis page shows GLM-5.2 (max) at 51, DeepSeek V4 Pro (max) at 44, Kimi K2.6 at 44, MiMo-V2.5-Pro at 42, and MiniMax-M2.7 at 38.
A widely circulated claim that M3 "scores 55" traces to Artificial Analysis's article of June 8, 2026 — published a week after launch, before the weights shipped — which states "MiniMax-M3 scores 55 on the Artificial Analysis Intelligence Index," placing it "just ahead of open weights peers Kimi K2.6 (54) and MiMo-V2.5-Pro (54)." That article names no index version, but it also put MiniMax-M2.7 at 50 where the live v4.1 page now shows 38, so it predates v4.1. Scores across Artificial Analysis index versions are not comparable, and the live page now reads 44.
arena.ai (the former LMArena, rebranded January 28, 2026, and a different Elo scale from Artificial Analysis) places MiniMax-M3 at rank 54 on the Text Arena leaderboard with 1447±6 Elo, and rank 16 with 1501 (+10/−10) on Code Arena's WebDev board. That is a web-development board, not a general coding board: arena.ai/leaderboard/code resolves to WebDev, and Code Arena also publishes an Image-to-WebDev board. The gap between a middling text rank and a strong WebDev rank is consistent with MiniMax's own positioning of M3 as a coding and agentic model rather than a general chat model.
On the MSA speedup numbers
Two different sets of speedup figures circulate, and they measure different things. The model card claims "9× prefill and 15× decode speedups compared to M2 at 1M context, reducing per-token compute to 1/20." The arXiv paper reports, "on a 109B-parameter model with native multimodal training," a "28.4x" reduction in per-token attention compute at 1M context and "14.2x prefill and 7.6x decoding wall-clock speedups" on H800 GPUs. The paper's numbers are not M3's numbers.
Limitations
- Not open source. The MiniMax Community License gates commercial use above $20M annual revenue behind prior written authorization and mandates "Built with MiniMax M3" attribution. Redistribution and commercial deployment are not free of conditions.
- A deliberate openness regression. M2 shipped under MIT in October 2025; M2.7 and M3 do not. Teams that adopted the M-series on permissive terms cannot assume continuity in future releases.
- All headline benchmarks are vendor-reported. SWE-Bench Pro 59.0, MCP Atlas 74.2, and the rest come from MiniMax. This page found no independent replication; the model card ships its comparison chart as a JPEG, and the repo's
.eval_resultsYAML only transcribes that chart. - Third-party scoring is far less flattering than the launch narrative. Artificial Analysis v4.1 places M3 at 44 — tied with DeepSeek V4 Pro and Kimi K2.6, and seven points behind GLM-5.2 (max) at 51.
- Long context is not free. Above 512K input tokens the API price doubles to $0.60 / $2.40 per million tokens, so the 1M window costs more than the headline rate implies.
- 428B parameters is a serious self-hosting bill. At bfloat16 the weights alone are roughly 854 GB; sparse activation reduces compute per token, not the memory needed to hold the experts.
- No published knowledge cutoff or training-data disclosure. Ground time-sensitive queries with retrieval or tool use.
- Absolute agentic scores remain low. SWE-fficiency at 34.8% and KernelBench Hard at 28.8% — MiniMax's own numbers — show long-horizon autonomy is unsolved.
Pricing & Access
MiniMax publishes a USD rate card on its pay-as-you-go pricing page. It labels the current rates "Permanent 50% off" list.
| Tier | Input (per 1M) | Output (per 1M) | Cache read (per 1M) |
|---|---|---|---|
| Standard, ≤ 512K input tokens | $0.30 | $1.20 | $0.06 |
| Standard, > 512K input tokens | $0.60 | $2.40 | $0.12 |
| Priority, ≤ 512K input tokens | $0.45 | $1.80 | $0.09 |
| Priority, > 512K input tokens | $0.90 | $3.60 | $0.18 |
MiniMax's launch blog also advertises subscription token plans: Plus at $20/month (~1.7B tokens), Max at $50/month (~5.1B tokens), and Ultra at $120/month (~9.8B tokens).
Self-hosting: weights are downloadable from Hugging Face via hf download MiniMaxAI/MiniMax-M3, but see the license section — this is not an unconditional grant.
Third-party hosting: OpenRouter lists minimax/minimax-m3 at 1,048,576-token context, matching MiniMax's $0.30 / $1.20 rate. As of this writing OpenRouter's live catalogue carries 8 MiniMax models — M3, M2.7, M2.5, M2.1, M2, M2-her, M1, and MiniMax-01.
Adoption signal: Hugging Face reports 233,589 downloads over the trailing 30 days for MiniMaxAI/MiniMax-M3 and 1,307 likes; the companion MiniMaxAI/MiniMax-M3-MXFP8 quantised repo adds 781,410. The older, text-only MiniMax-M2.7 still draws 1,181,781 downloads over the same window — even combined, M3's two official repos have not yet displaced it.
Ecosystem & Tools
- MiniMaxAI/MiniMax-M3 on Hugging Face — weights, model card, and the MiniMax Community License text
- MiniMax-AI/MiniMax-M3 on GitHub — repository created June 1, 2026
- MiniMax-AI/MSA — the standalone MiniMax Sparse Attention operator
- arXiv:2606.13392 — "MiniMax Sparse Attention," the technical report, submitted June 11, 2026
- MiniMax Agent — MiniMax's own agent product running on M3
- MiniMax Platform docs — API reference and release notes
- vLLM, SGLang, Transformers, KTransformers, unsloth — the five serving paths MiniMax documents; Transformers exposes the model as
minimax_m3_vl - NVIDIA deployment guide — Blackwell serving notes, BF16/MXFP8 precision
Community & Resources
- MiniMax M3: Frontier Coding, 1M Context, Native Multimodality — the official launch blog, June 1, 2026
- MiniMax API release notes — the authoritative release-date list for the M-series
- MiniMax-M3 on Artificial Analysis — Intelligence Index v4.1 score of 44
- MiniMax-M3: Leading open weights model, once the weights are released — Artificial Analysis, June 8, 2026; its "55" score predates Intelligence Index v4.1
- MiniMax Drops State-of-the-Art AI Agent Model — Then Quietly Changes the License — Decrypt, April 13, 2026, on the M2.7 license change
- MiniMax doubles in Hong Kong debut — CNBC on the January 9, 2026 HKEX listing
- MiniMax also builds Hailuo 2.3, its video generation model, released October 28, 2025
- Compare with DeepSeek V4, Kimi K2.6, GLM-5.2, MiMo-V2.5-Pro, and Qwen3.7-Max