Model releases.

Generates video with synced audio offline, using a 32B vision-language text encoder.

Generates video with synced audio, clips up to 30 seconds, and strong scene consistency.

Generates video with synchronized audio, multi-reference consistency, and clip editing.

A faster open-weight reasoning model that reads text, images and audio.

Erases what a mask marks from an image and fills in the background.

Transcribes recorded audio to plain text, using hints for names, with no timestamps.

Video from text, image or video, with synced audio, lip-sync and a cheap draft mode.

The highest-fidelity MAI-Image 2.5 tier, for photorealistic and artistic images.

Generates and edits images with balanced visual quality, layout fidelity, and speed.

Generates images with dense layouts, small-text rendering, and multilingual visual design.

Efficient multimodal model with stronger reasoning, coding, and computer-use quality.

A 2.8T open-weight reasoning model for agents and coding, with 1M context.

An open-weight reasoning model for coding and agents that reads images and audio.

Fast, affordable text model for high-volume, latency-sensitive, low-budget tasks.

A top-tier text model built for frontier coding, long-horizon agentic tasks, and research.

Generates images up to 2K with strong prompt fidelity and multi-reference editing.

Plans layouts and calls tools before rendering for faithful, precise image editing.

A small vision model that answers questions about images, captions them and reads text.

Agentic coding model with 33B mixture-of-experts, 3B active, in full BF16 precision.

A 33B MoE model for agentic coding and long software tasks with tool use.

Handles agentic coding tasks with a 33B mixture-of-experts model, 3B active.

A fast, low-cost text model for coding and tool use with near frontier performance.

Reads a script or builds a full audio scene with voice, music and sound effects.

Fast, low-cost image generation making 1K images in about 4 seconds for high volume use.

Agentic coding with 256K context, an open-source model tuned to keep improving itself.

Agentic coding model with 256K context and native tool use, self-improving, MIT licensed.

Fast reasoning model with a 256K context, supporting tool calling and image input.

Text-to-speech with inline emotion cues, multi-speaker synthesis, and voice cloning.

The S2.1 Pro model at no charge for testing, without latency guarantees.

Generates or edits clips up to 4K from text, frames, images or a source video.

Generates 3 to 15 second clips up to 1080p with native audio from text or a frame.

Generates video from text, animates images, or uses up to nine reference images.

Turns a still image into a short video clip with synchronized audio.

A coding model with 1M-token context and deep reasoning for end-to-end engineering work.

Generates video at 480p or 720p with synced audio, about half the cost per second.

Handles demanding reasoning, coding, and agentic work with vision and 128K output.

A compact code model tuned for fast local code generation, completion, and refactoring.

A 12B multimodal model with strong text and image understanding.

Generates photorealistic and artistic images from a prompt or one reference image.

Edits 2 to 30 second video clips from a text prompt, with keyframe and outpainting.

A frontier model for advanced reasoning, coding, and agentic tool use with vision.

A compact 8B model tuned for long-context reasoning that runs efficiently on-device.

Generates brand-consistent images with strong control over typography, layout, and colour.

Creates higher-resolution, print-ready visuals for marketing and packaging design.

Strong reasoning, coding, and tool use, with vision and real-time knowledge from X.

A 30B dense model built for strong reasoning in business, coding, and multilingual tasks.

A compact 3B instruct model for multilingual chat, tool use, and JSON output.

An 8B dense model for instruction following, RAG, and code tasks.

Handles reasoning tasks with tool calling and image input at a lower cost than Pro.

A compact, low-cost model for fast everyday tasks, with tool calling and image input.