Model Square
This site currently has 93 models enabled.
A 1.6-trillion-parameter model with ultra-long context, strong reasoning, coding and long-document processing.
The official DeepSeek V4 Pro release with stronger agent capabilities than the preview.
A lightweight, balanced omni-modal model for text, image, audio and video understanding with cross-modal reasoning.
A next-generation flagship with major gains in coding, reasoning, multimodality and long-context understanding.
Designed for agent scenarios such as OpenClaw, combining fast reasoning with strong performance.
Built for long-horizon work, autonomously planning, executing, optimizing and delivering engineering results for up to eight hours.
A one-million-token context model with leading open-source performance on long-horizon tasks.
An open 3-trillion-parameter model with a one-million-token context for long-horizon coding, knowledge work and reasoning.
A flagship model combining leading coding, agent tool use, search and office performance with high speed and low cost.
A dual-thinking vision model with stronger visual coding, spatial perception, multimodal reasoning and long-video understanding.
A native multimodal open flagship with 397B total and 17B active parameters, balancing performance and cost.
A native vision-language Flash model improved for agentic coding, math and code reasoning, spatial intelligence and detection.
A text preview model improved for vibe coding, coding agents, frontend development and long-tail knowledge.
A native vision-language Plus model with stronger coding, multimodal recognition, OCR and object localization.