Switching between “Think” and “Non-Think” modes feels like juggling two interns: one who needs constant hand holding or the whole workflow stalls, and the other who just chugs along smoothly without much babysitting. How do you make sure the mode switch doesn’t kill the flow and stays seamless, so users don’t end up frustrated juggling interruptions?
Congratulations on the launch! DeepSeek-V3.1 sounds like a really exciting step toward the agent era—especially the hybrid design of thinking and non-thinking modes. I’m curious, in what kinds of real-world use cases do you think the switching between modes will shine the most?
Love the hybrid thinking/non-thinking mode—switching between those on the fly is super smart, tbh. Gonna be wild to see what people build with this!
I have tried version 3.1 of DP and it is indeed smarter.
But it seems that the API price has also increased, although it is still very cheap!
Hey Hunters 👋
I am excited to hunt DeepSeek-V3.1 — our first step toward the agent era!
✅ Hybrid inference: Think & Non-Think — one model, two modes
✅ Faster reasoning: DeepSeek-V3.1-Think answers quicker than R1-0528
✅ Stronger agent skills: Improved tool use + multi-step task handling
✅ 128K context for both chat & reasoner modes
✅ Strict Function Calling support in Beta API
✅ Better performance on SWE/Terminal-Bench + complex reasoning tasks
✅ Open-source weights available for the community
You can even toggle between Think & Non-Think modes via the new DeepThink button.
We’d love your thoughts as we continue building toward smarter, faster, and more capable AI agents 🚀