AI
New #1 open-source AI model is here! GLM 5.2AI Search17 Jun 2026
Mentioned as a multimodal competitor to GLM, which lacks native vision.
Open-weights native multimodal MoE model from MiniMax, ~428B parameters (~23B active) with 1M-token context.
Updated EU status checked Last shown
MiniMax M3 at a glance
- What is MiniMax M3?
- MiniMax M3 is a model from MiniMax. Open-weights native multimodal MoE model from MiniMax, ~428B parameters (~23B active) with 1M-token context.
- Who makes MiniMax M3?
- MiniMax M3 is made by MiniMax.
- Does MiniMax M3 work in the EU?
- Available in the EU. Weights are downloadable from Hugging Face with no EU exclusion visible in the provided model card; the custom "minimax-community" license text was not reviewed here. Hosted API and MiniMax Code availability for the EU was not verified. Last verified 14 Nov 2026.
- How much does MiniMax M3 cost?
- Open source. Weights free to download under the MiniMax community license; hosted access through the MiniMax API (pricing not given in the provided material) and MiniMax Code. Licence: MiniMax Community License.
- Where was MiniMax M3 shown?
- In 4 creator videos by AI Search, Matt Wolfe. Most recently by AI Search on 17 Jun 2026, at 21:38 in “New #1 open-source AI model is here! GLM 5.2”.
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
About
MiniMax M3 is a multimodal mixture-of-experts model from MiniMax with roughly 428B total and 23B activated parameters and a 1M-token context window. According to the model card, it is trained on mixed modalities from the first step, covering text, image and video input, and is aimed at coding, agentic and "cowork" tasks.
It uses MiniMax Sparse Attention (MSA), a sparse attention operator for million-token contexts. MiniMax reports 9x prefill and 15x decode speedups over M2 at 1M context. The model supports three reasoning modes via a thinking parameter: enabled, adaptive and disabled. A technical report is on arXiv (2606.13392).
Weights are on Hugging Face under the MiniMax community license (not an OSI open-source license) and can be served with SGLang, vLLM, Transformers, KTransformers and unsloth. It is also available through the MiniMax API and the MiniMax Code agent product. A YouTube creator mentioned a smaller FP8 version and placed it among the leading open models on Artificial Analysis; this was not verified here.
Key features
- ~428B total / ~23B active parameter MoE
- 1M-token context via MiniMax Sparse Attention
- Native multimodal input: text, image, video
- Three reasoning modes: enabled, adaptive, disabled
- Downloadable weights; served via SGLang, vLLM, Transformers
- Focus on coding and long-horizon agentic tasks
4 moments
As seen on YouTube
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
AI
RIP Claude Fable, open-source AI unleashed, full body avatars, new Google models, new TTS: AI NEWSAI Search14 Jun 2026
Open-weights model with 427B parameters (23B active) and 1M token context, leading open models on Artificial Analysis. Weights released this week, with a smaller FP8 version.
AI
Full body waifus, AI dreams, realtime AI music, open-source Gemini Omni: AI NEWSAI Search7 Jun 2026
Upcoming open-source multimodal model with 1M context, sparse attention and strong agentic coding, beating GPT-5.5 on SWE-Bench Pro. Leads open models on Artificial Analysis but lags GLM 5.1 on Arena; weights due in 1-2 weeks.
MA
AI News: Microsoft Finally Reveals Their Plan!Matt Wolfe5 Jun 2026
Coding model with 1M token context, claims to beat GPT 5.5 and Gemini 3.1 on SWE-Bench Pro, and is inexpensive.
Keep exploring
Similar AI tools
Closest neighbours by meaning, not just by category.
IS
Inkling Small
Model · Thinking Machines
Open-weights multimodal MoE model (276B total, 12B active) from Thinking Machines; text, image and audio in, text out.
Available in EULanguage Models
27 Jul 2026Open source
L2
LongCat 2.5
Model · Meituan
Meituan's open-weights 1.6T-parameter multimodal MoE model (48B active) for agentic coding and visual tasks.
Available in EULanguage Models
4 Nov 2026Open source
Q3
Qwen 3.8 Max
Model · Alibaba
Alibaba's open-weights 2.4T-parameter MoE language model (95B active) from the Qwen3.8 family, with a hosted Max version.
Available in EULanguage Models
Release date n/aOpen source
LI
Ling-3.1-flash
Model · inclusionAI
Open-weights 560B hybrid-linear MoE reasoning model (25B active, 1M context) from inclusionAI, MIT licensed.
Available in EULanguage Models
13 Oct 2026Open source
O1
Ornith 1.5
Model
Open model family (9B dense, 35B MoE, 397B MoE) trained with a self-improvement loop for reasoning, coding and agentic tasks.
Available in EULanguage Models
Release date n/aOpen source
L2
LongCat 2.0
Model · Meituan
Meituan's open-weights 1.6T-parameter mixture-of-experts LLM (48B active), aimed at agentic coding and reasoning tasks.
Available in EULanguage Models
Release date n/aOpen source