DeepSeek-V3.1
Our first step toward the agent era
Visit website →About
DeepSeek-V3.1 is a large open-weight language model with a hybrid design: one model serves both a step-by-step thinking mode for hard reasoning and a direct mode for fast everyday answers, switched by a template rather than a separate model.
It improves on earlier DeepSeek releases in tool use and multi-step agent tasks, supports a long context window, and keeps the low API pricing the lab is known for. Weights are published for self-hosting.
It is aimed at developers building agents, coding assistants, and analysis pipelines who want strong reasoning without frontier-lab API costs, and at researchers who need an openly available model they can run and fine-tune themselves.
AI-assisted draft, human-reviewed before publishing — see how we choose & review tools.
Why we picked it
DeepSeek-V3.1 earns its spot on price and openness — an open-weight model that switches between a reasoning mode and a fast mode by template, priced well under closed alternatives, worth it for anyone building on top of a model rather than just chatting with one.
Related tools
- 1 BOSS Console A tiny OS that gives always-on AI agents a place to work with real tools Free
- 2 withHuman Human approval layer for AI agents built on an open protocol Freemium
- 3 guilds.run Open-source AI agent teams you can actually inspect Free
- 4 OpenSwarm An AI-first OS where agent swarms do the work in parallel Free
- 5 Opengeni Open-source infrastructure for AI agents in your product Freemium
- 6 iFixAi Independent auditing for AI agents that catches hidden misalignment Freemium