Model releases.

Decision model with a context window of over 500K tokens.

Handles coding, agentic tasks, and knowledge work, with vision and 500K-token context.

A high-speed variant of the flagship coding model with faster output at a higher price.

Generates and edits images from up to 10 references, with optional prompt rewriting.

9B decision model for classification, triage and rubric scoring.

Compact multilingual decision model that answers in a single forward pass.

Transcribes multi-speaker conversations with speaker labels and emotion cues.

Transcribes speech with word timings and speaker labels, guided by key-term hints.

An omni-modal reasoning model for agentic work that reads text, images, audio and video.

A ternary-weight vision model with tools and thinking, about half a 4-bit model's size.

Turns text or first and last frames into 5 to 15 second clips up to 768p with audio.

A fast, low-cost image model for high-volume generation and multi-reference editing.

Writes studio-grade songs with vocals or instrumentals from a text prompt.

Transcribes clinical and medical speech, with word-level timing and speaker labels.

Edits a source clip from a text prompt, keeping its length, aspect ratio and audio.

Turns text, images, or audio into 20-second clips with native audio.

The fastest H3 tier, making 480p clips with audio from text, frames or references.

Generates design-ready images and edits from up to five reference images.

A faster, lower-cost image model that also edits from up to five reference images.

Generates full-length songs with vocals or instrumentals from a prompt, lyrics or images.

Transcribes speech in 60 languages with auto language detection, returning plain text.

Handles complex reasoning, coding, research, and agent work with a 1M-token context.

Flagship tier that reasons longer on each answer for the highest quality on hard tasks.

Edits a source clip from a text prompt, with optional reference images and a draft mode.

A flagship reasoning model for coding and agent work, with image input and 1M context.

The most capable Flash model for software engineering and agentic tasks, with 1M context.

A multimodal reasoning model for long agentic and coding workflows, with 1M context.

A faster, lower-cost H3 Max tier for text and frame-guided clips with native audio.

Transcribes push-to-talk and speaker-aware speech into plain text.

Generates fast video clips with native synced audio and first-to-last frame control.

A fast, low-cost multimodal reasoning model for coding and visual understanding.

Transcribes speech to plain text with automatic language detection.

An efficient multimodal model for coding and long agent tasks, with image input.

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

Compact 3B instruction-tuned model with multilingual support and tool use.

An 8B dense model for instruction following, RAG, and code tasks, with native tool use.

Generates video with native audio, offering lower latency and a 30-second ceiling.

Upscales a source clip by 1.5 to 3 times, faithfully or with added fine detail.

A 35B MoE model with 256K context for reasoning, coding, and agentic tool use.

9B self-improving model with 256K context for reasoning, coding, and agentic work.

A v3 voice model tuned for natural dialogue, in more than 70 languages.

A 27B multimodal model with 256K context for coding, research, and long agentic tasks.

A flagship open-weights coding model with 1M context for long-horizon agentic work.

V4 model with 1M context, dual thinking modes, and stronger agentic performance.

Generates video with synced audio and first-to-last-frame control, up to 20s at 4K.

Generates high fidelity video with synced audio and directable camera moves up to 1080p.

A 30B mixture-of-experts model with 3B active parameters for always-on agents.

A 30B multimodal model for always-on local agents, tuned for tool use and long tasks.

Generates and edits images with quality-aware output control at up to 2K resolution.

Generates clips up to 30 seconds with native audio from text, images, or references.