MiniMax M3: open weights land for the 428B multimodal model with 1M context
MiniMax published the open weights of M3, a 428B-total / 23B-active natively multimodal MoE with a 1M-token context, after its June API launch.
Why it matters
The weights drop turns June's API-only flagship into a downloadable model: 1M context, native text-image-video training from step zero, and a 12-hour autonomous ICLR paper replication demo, now runnable locally via vLLM and llama.cpp.
What you could build with it
A startup building a long-document research assistant could run M3 locally with its 1M-token context to ingest whole codebases or paper collections in one window, avoiding per-token API bills while keeping data on its own cluster.
Does it hold up?
Usable now through the hosted API and community quants: Unsloth's 2-bit GGUF runs on roughly 150GB of RAM, but full weights are 428B, and the MiniMax Community License requires written authorization for companies over $20M in yearly revenue.
Built with MiniMax M3
- MiniMax unveils M3: an open-weight model combining 1M context, native multimodality and frontier codingarticle · AI China News covers the M3 launch: BrowseComp 83.5, the 12-hour ICLR replication, the 147-iteration CUDA kernel optimisation, and integrations with Claude Code, Cline, Cursor and others.
- MiniMax M3 reviews: what 1.1k people sayarticle · Community verdict aggregator: M3 ranked #85 of 318 assistants with a 6.2 vibe score from 1,115 voices as of Nov 5.
- 10 Best Open Source LLMs in 2026, by Benchmark and Licensearticle · Respan's roundup lists M3 at 31.2 on its index with the MiniMax Community License terms, noting commercial use needs written authorization above $20M yearly revenue.
- vllm-project/vllmgithub · 96,387 ★ · vLLM v0.33.0 registers MiniMaxM3SparseForConditionalGeneration, so M3 runs on the standard open serving stack.
- ggml-org/llama.cppgithub · 133,440 ★ · llama.cpp release b11861 adds the minimax-m3 architecture, enabling local GGUF inference paths.
Learn more
licence and weights explainer →
First spotted on article: source.
More AI models
VoiceStudio: open-source, fully-local ElevenLabs alternative for voice cloning, dubbing, and audiobooksdebpalash/VoiceStudio is a desktop app that runs voice cloning, voice design, video dubbing, dictation, transcription…models · JEV 0.75Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73NVIDIA Isaac Lab 3.0 GA + Isaac OS open runtime + Agile ONE SN humanoidMajor Isaac platform expansion at GTC Berlin Oct 21: Isaac Lab 3.0 is generally available as the open-source robot…models · JEV 0.73NVIDIA Nemotron 3.5 Retriever collection: Embed 8B/1B, Rerank 8B, Parse 631MNVIDIA's Oct 21 collection drop for agentic retrieval: Nemotron-3.5-Embed in 8B BF16, 1B BF16 and 1B NVFP4…models · JEV 0.72Kimi K3.1 Pro: Moonshot's 2.8T open-weight flagship officially launchesMoonshot AI officially launched Kimi K3.1 Pro on Oct 23, a mixture-of-experts model with 2.8 trillion total / 104…models · JEV 0.72NVIDIA Cosmos 3.5 world models: now openly availableNVIDIA's Cosmos 3.5 family of open world foundation models for robots, autonomous vehicles and vision agents is now…models · JEV 0.71
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.