Model releases.

An open-weight decision model for choice, score and yes/no answers on text and images.

The largest Drex decision model, with a 262K-token context.

Answers choice, yes/no and score questions with probabilities, never free text.

Text-to-speech you direct in plain language, with sounds, pauses and several speakers.

A flagship reasoning model for software engineering, science and agents, with 1M context.

A low-cost reasoning model with 2.45B active parameters, built for tool-calling agents.

A 35B mixture-of-experts decision model, the larger successor to Drex v1.5.

A frontier reasoning model for complex agentic tasks, with a 1M-token context.

The most accurate d3, answering questions about text or images.

The smallest d3, for fast local routing and classification of text or images.

The general-use d3, answering questions about text or images.

A larger d3 for harder decisions about text or images.

Low-cost 8-step image generation and editing from up to 10 reference images.

Open decision model that reads text and images and returns typed decisions.

Text-to-speech with preset voices, and voice cloning on Runware.

Small decision model that returns a calibrated probability for each answer option.

A 2B decision model for classification, routing and scoring on Apple silicon Macs.

A faster, low-latency variant of Solar Decide with a context of over 500K tokens.

A fast, low-cost model for classification, extraction and routing with a 1M-token context.

A tiny offline decision model for choice, score and yes/no answers on text and images.

Generates and edits images with sharper composition and steadier subjects, up to 4K.

Open-weight multimodal model for reasoning, coding and agent work with a 512K context.

Perplexity's 27B decision model, a newer checkpoint of Decider V1 that reads a state and returns typed, probabilistic answers to choice, score and yes/no questions.

A 3B pixel-space image model that generates from prompts on your own machine.

Generates and edits images with up to ten references, sharp text, and native 4K output.

Open 27B decision model that turns a state and typed questions into decisions.

Fast open 9B decision model for typed classification and routing.

Reads text aloud in 23 languages with detailed prosody and a consistent speaker.

Reads text aloud in 23 languages with low latency for real-time use.

A low-cost model for short clips with audio from text or a starting image, up to 1080p.

Generates 5 to 15 second clips with sound from text, a starting image or references.

Generates and edits images with accurate typography and up to five references.

Transcribes speech to plain text; long recordings go out in two-minute pieces.

Liquid's structured decision model, served as a System One endpoint for choice, score and yes/no answers.

Updated high-end tier for agentic coding and document work, with vision and 1M context.

Works longer on each answer for higher quality results on hard text tasks.

A fast everyday model for coding and tool use, with vision and a 1M-token context.

The most expressive speech model, with audio tags and more than 90 languages.

A faster tier of v4 with the same audio tags at half the per-character price.

A 9B decision model that returns a calibrated probability for each answer.

Smaller experimental decision model, a 0.8 GB download.

A higher-throughput variant of the flagship reasoning model, at a higher price.

Experimental 4B decision model for routing, triage and policy checks.

Handles long agentic coding and knowledge work with vision and 1M-token context.

Text-to-speech model for long-form narration in 130 languages with vocal cues.

A fast, low-cost text-to-speech model for narration and voice agents in 101 languages.

Fast, low-cost tier for high-volume chat, classification, and light agentic work.

Answers hard questions by reasoning longer on each one to give higher quality results.

Cost-efficient high-end tier for demanding coding and agentic work, with 1M context.

Handles hard text tasks by reasoning longer for higher quality answers.