Search the full model catalog.

A dense 9-billion parameter language model optimized for chat and reasoning tasks. Designed for efficient deployment and high-quality responses.

A decensored build of Qwen3.5-27B, made with Heretic, a tool that edits the model's weights to cut refusals. Text chat with an optional thinking mode and function calling.

SongGeneration 2 generates full songs with vocals and instrumentals from tagged lyrics, up to 4 minutes 30 seconds long, styled by caption tags or a reference clip.

Writes an instrumental track for a video and mixes it with the clip's own audio. Describe the music in a caption, or leave it empty and the app picks a style from the footage.

A lightweight ByteDance language model for chat and tool calling, with image understanding and a 256K-token context window.

Turns a start image, and optionally an end image, into a clip of 6 to 20 seconds with its own soundtrack using LTX-2.3, the Lightricks update to LTX-2 with a redesigned VAE and native portrait video.

Open-source English TTS model from Resemble AI, a lighter 350M-parameter Chatterbox variant. Clones a voice from a short clip and performs inline tags like [laugh], [chuckle] and [cough].

Hunyuan3D-2.1 is Tencent's open image-to-3D model: one picture in, a GLB mesh out, with optional PBR materials (base color, metallic, roughness) for game engines and 3D tools.

A real-time voice conversation tool using ElevenLabs' AI voice agents. Customize voices, behaviors, and languages for interactive AI experiences.

GPT Mini Realtime enables low-latency, bidirectional streaming for voice and text. Build interactive, responsive AI experiences that feel natural and immediate.

GPT Realtime enables low-latency, bidirectional streaming for voice and text. Build interactive, responsive AI experiences that feel natural and immediate.

GLM-4.7-Flash is a 30B-A3B mixture-of-experts language model from Z.ai. It writes out its reasoning before answering, and Z.ai recommends it mainly for frontend and backend coding.

Z.ai's GLM-Image creates and edits pictures by pairing a 9B autoregressive generator with a 7B diffusion decoder. It is built for images that carry a lot of text, such as posters and labeled diagrams.

Pulls the text out of an image and translates it, using Google's 12B TranslateGemma and a choice of 54 languages. Google reports it beats the original Gemma 3 27B on text (WMT24++), though not on images (Vistra).

NVIDIA's 600M-parameter English speech-to-text model, with a cache-aware streaming encoder and an RNN-T decoder. On Wiro it takes one recording and returns the transcript as plain text.

The undistilled 9B base model of FLUX.2 [klein], for text-to-image and editing with up to 4 reference images, aimed at research and fine-tuning under a non-commercial license.

Single-speaker text-to-speech from Microsoft Research, built on Qwen2.5-0.5B and designed for streaming text input. It reads scripts of up to about 10 minutes, primarily in English, with 25 preset voices.

Places one portrait in a birthday or zodiac-season scene and writes your caption into the set, such as an age in gold balloons or a sign name as the title, then animates it.

Turns a static logo into a short animated video using one of 36 scene presets, such as a Times Square billboard, latte art, a Parisian storefront or a neon bar sign.

Meituan's instruction-based image editor. You describe a change in English or Chinese, and it applies that change while aiming to keep everything you did not mention as it was.