What is DeepSeek-V3.2?
DeepSeek's open, MIT-licensed reasoning model for agents, with thinking during tool calls.
Who is DeepSeek-V3.2 best for?
DeepSeek-V3.2 is best for teams that self-host an open model for agents that reason while calling tools.
Self-hosted agents
It was DeepSeek's first model to reason inside tool calls, and it can use tools in both thinking and non-thinking modes.
Long-context workloads
DeepSeek Sparse Attention is designed to cut the compute cost of long inputs while keeping the model's performance.
Commercial products on open weights
Both the code and the weights are under the MIT license, so the model can be built into your own products.
Reasoning research
DeepSeek published a technical report and its olympiad submissions so others can check the results.
How much does DeepSeek-V3.2 cost?
DeepSeek-V3.2 is free to download under the MIT license, and DeepSeek's own API no longer serves it, since the API has moved to its V4 models.
- Download from Hugging Face
- About 685B parameters
- Same local setup as DeepSeek-V3.2-Exp
DeepSeek's API pricing page now lists only DeepSeek-V4.1-Flash and DeepSeek-V4-Pro, at $0.30 and $1.32 per million input tokens and $1.20 and $3.96 per million output tokens at peak hours, half that off-peak. To use V3.2 itself, you host the open weights. Full pricing on api-docs.deepseek.com. We checked these prices on .
How do readers rate DeepSeek-V3.2?
No one has rated DeepSeek-V3.2 yet. Be the first.
Sign in to rate. One rating per account.
Is DeepSeek-V3.2 open source?
Yes. DeepSeek-V3.2's code is public under the MIT licence, so you can run it on your own server.
What are the best DeepSeek-V3.2 alternatives?
DeepSeek-V3.2 is an open source alternative to Kimi K2.5, MiniMax M2.5, Gemini 3.1 Pro and GPT-5.2. Step-3.5-Flash and MiMo-V2-Flash do the same job and are open source too.
Moonshot's open multimodal model for visual coding and agent swarms of up to 100 sub-agents.
MiniMax's open-weight coding and agent model, at $0.30 per million input tokens.
Google's advanced reasoning model for coding, research and agents.
API-only GPT-5.2 with a 400K context and reasoning controls.
StepFun's 196B MoE with a 256K context for coding and agents.
Xiaomi's 309B MoE with a 256K context, open for download.
Questions about DeepSeek-V3.2
What is DeepSeek-V3.2?
DeepSeek's open, MIT-licensed reasoning model for agents, with thinking during tool calls.
Who is DeepSeek-V3.2 for?
DeepSeek-V3.2 is best for teams that self-host an open model for agents that reason while calling tools.
Is DeepSeek-V3.2 free?
Yes. DeepSeek-V3.2 has a free Open weights plan: download from Hugging Face.
Is DeepSeek-V3.2 open source?
Yes. DeepSeek-V3.2's code is public under the MIT licence, so you can run it on your own server.
What are the best DeepSeek-V3.2 alternatives?
The closest alternatives on Recaply are Kimi K2.5, MiniMax M2.5, Gemini 3.1 Pro, GPT-5.2, Step-3.5-Flash and MiMo-V2-Flash.
Where can I try DeepSeek-V3.2?
On its official website, huggingface.co.
Can I still use DeepSeek-V3.2 through DeepSeek's API?
No. DeepSeek's API now offers only DeepSeek-V4.1-Flash and DeepSeek-V4-Pro, so DeepSeek-V3.2 is available only as open weights on Hugging Face, which you run with the same setup as DeepSeek-V3.2-Exp.







