Image and video understanding
M3 was trained as a native multimodal model and accepts image and video input. The first Dreammmer integration will stay focused on one image plus text.
Next-model release tracker
MiniMax M3 Pro Online: Release Status and M3 Experience
Follow the reported MiniMax M3 Pro and possible MiniMax M4 while exploring what the official M3 already confirms: native multimodality, 1M context, coding, agents, and three reasoning modes.
Dreammmer is independent from MiniMax. MiniMax has not officially announced M3 Pro or M4. Media reports describe an unnamed 2.7-trillion-parameter model and say M3 Pro is an internal working name; the preview below stays disabled and sends no text or image.
MiniMax multimodal workspace
Planned initial runtime: MiniMax-M3 · thinking=adaptive
Reasoning mode
Think
Always reason
Adaptive
Model decides
Fast
Thinking off
Optional image
M3 understands image input; it does not generate the output image here.
Text and one-image input will open only after a supplier model ID, request format, limits, and moderation path are verified.
Prepared tasks
Analyze a screenshotReview codeSynthesize long materialPlan a multi-step task
A future live response will show the actual model ID and thinking mode. Model capabilities and tools supplied by this page will remain clearly separated.
Model response
After a provider is connected and verified, real M3 responses will appear here. This preview does not invent a MiniMax M3 Pro response.
Confirmed M3 capability baseline
What users can reasonably expect to test
These jobs come from MiniMax's official M3 release and model card. They are a baseline for the next model, not a claim about an unreleased version.
Coding and debugging
MiniMax positions M3 for bug fixes, frontend and backend development, performance work, and longer collaborative engineering sessions.
One-million-token context
M3 documents a 1M context window for large documents, codebases, logs, and long-running tasks, powered by MiniMax Sparse Attention.
Agentic and computer tasks
M3 is trained for multi-step agent work and computer use. Execution still requires real tools and permissions supplied by the application.
Names, models, and modes
Separate confirmed versions from reported names
M3 Pro and M4 are search terms around the next release. M3 and M2.7 are official models. The table keeps those different evidence levels visible.
NameCurrent statusWhat it means
MiniMax M3 Pro
Reported internal name
The Information says the 2.7T model is known internally as M3 Pro, but the public release name may change.
MiniMax M4
Unconfirmed search name
A natural numerical successor to M3, but no MiniMax source currently announces an M4 model.
MiniMax M3 · Think
Official · thinking=enabled
Keeps reasoning on for complex coding, research, and agent tasks.
MiniMax M3 · Adaptive
Official · thinking=adaptive
Lets M3 decide when extra reasoning is useful and is the planned default for this page.
MiniMax M2.7
Official previous generation
An earlier open model in the M-series. M3 adds native multimodality and a larger 1M context baseline.
This URL will remain the MiniMax M3 Pro canonical page. If MiniMax publishes another final name, Dreammmer will update verified facts and connect the official name without presenting a rumor as a release.
Verified release picture
What is known as of July 20, 2026
Official MiniMax material supports M3 facts. Reuters and The Information are shown separately for the reported next model.
Last verified: July 20, 2026
Official status
MiniMax's model site and release pages list M3 as the current flagship. No official M3 Pro or M4 announcement, model ID, pricing, or release date was found.
Next-model report
Reuters reports that MiniMax is developing a 2.7-trillion-parameter open-weight model that could arrive as early as Q3 2026; MiniMax declined to comment.
Current M3
M3 launched on June 1, 2026 with about 428B total and 23B active parameters, native image and video understanding, 1M context, coding, agents, and computer use.
API and weights
MiniMax says the M3 API is available, and official weights are published on Hugging Face. Dreammmer has not yet verified a provider integration, so this page sends no request.
Clear product boundaries
Use MiniMax to understand; use an image model to create
MiniMax M3 can understand visual input and return text. Dreammmer keeps image generation in its Seedream workflow so visitors know which model performs each job.
MiniMax M3 Pro questions
Release name, availability, M4, API, and capabilities
The short answers distinguish MiniMax's confirmed M3 facts from reporting about the next model.
Has MiniMax M3 Pro been released?
No official MiniMax M3 Pro release was found as of July 20, 2026. The Information reports M3 Pro as an internal working name for a 2.7T model, while MiniMax has not confirmed the report or final name.
Is the next MiniMax model called M4?
Not officially. M4 is a plausible numerical successor and a useful related search term, but MiniMax has not announced that name. Current reporting points to M3 Pro as an internal name that could still change.
Can I use MiniMax M3 Pro on Dreammmer now?
No. There is no confirmed public M3 Pro model ID or verified Dreammmer provider. The preview stays disabled and sends no prompt or image.
What model will Dreammmer connect first?
The first planned runtime is the official MiniMax-M3 model in adaptive thinking mode. Image input will be enabled only if the chosen provider documents and passes the full upload and moderation flow.
Does MiniMax M3 generate images?
The official M3 material describes image and video understanding with text output, not this page as an image generator. Dreammmer will use Seedream for image creation and MiniMax for analysis, coding, research, and planning.
Will this page be rebuilt after the model launches?
No. The URL and core sections will remain stable. Dreammmer will update the release notice, verified specifications, comparison, runtime, and API details after checking official MiniMax material and the connected provider.
Sources and verification
Official MiniMax sources support M3 capabilities and modes. Reuters and The Information support only the reported next-model size, timing, and working name.