What is DeepSeek-V3.2-Speciale?
DeepSeek's high-compute V3.2 variant for hard math and coding problems, released as open weights.
Who is DeepSeek-V3.2-Speciale best for?
DeepSeek-V3.2-Speciale is best for researchers who want an open model for olympiad-level math and competitive programming.
Olympiad-level math
DeepSeek reports gold-level results at the IMO and the CMO maths olympiads from this variant.
Competitive programming
It also reached gold-level results at the ICPC World Finals and the IOI, according to DeepSeek.
Checking reasoning claims
DeepSeek released the model's olympiad submissions alongside the weights so others can verify them.
How much does DeepSeek-V3.2-Speciale cost?
DeepSeek-V3.2-Speciale is free to download under the MIT license; its API access was temporary and has ended, so you run the weights yourself.
- Download from Hugging Face
- About 685B parameters
- Reasoning only, no tool calls
At launch DeepSeek served Speciale on a temporary API endpoint at DeepSeek-V3.2's price, and that endpoint has closed. DeepSeek's API now lists only DeepSeek-V4.1-Flash and DeepSeek-V4-Pro. Speciale uses more tokens per answer than V3.2. Full pricing on api-docs.deepseek.com. We checked these prices on .
How do readers rate DeepSeek-V3.2-Speciale?
No one has rated DeepSeek-V3.2-Speciale yet. Be the first.
Sign in to rate. One rating per account.
Is DeepSeek-V3.2-Speciale open source?
Yes. DeepSeek-V3.2-Speciale's code is public under the MIT licence, so you can run it on your own server.
What are the best DeepSeek-V3.2-Speciale alternatives?
DeepSeek-V3.2-Speciale is an open source alternative to Gemini 3.1 Deep Think and GPT-5.5. Command A+, Nomos 1, DeepSeek V4 Pro and NousCoder-14B do the same job and are open source too.
Parallel reasoning for maths, science, engineering and code, on Google AI Ultra.
OpenAI's agentic model for coding and professional work, with a Pro tier.
Cohere's open multimodal reasoning model for enterprise agents.
Nous Research's open maths reasoning model.
DeepSeek's 1.6T-parameter open MoE model for coding and agents, with a 1M-token context.
An Apache-licensed 14B model tuned for competitive programming.
Questions about DeepSeek-V3.2-Speciale
What is DeepSeek-V3.2-Speciale?
DeepSeek's high-compute V3.2 variant for hard math and coding problems, released as open weights.
Who is DeepSeek-V3.2-Speciale for?
DeepSeek-V3.2-Speciale is best for researchers who want an open model for olympiad-level math and competitive programming.
Is DeepSeek-V3.2-Speciale free?
Yes. DeepSeek-V3.2-Speciale has a free Open weights plan: download from Hugging Face.
Is DeepSeek-V3.2-Speciale open source?
Yes. DeepSeek-V3.2-Speciale's code is public under the MIT licence, so you can run it on your own server.
What are the best DeepSeek-V3.2-Speciale alternatives?
The closest alternatives on Recaply are Gemini 3.1 Deep Think, GPT-5.5, Command A+, Nomos 1, DeepSeek V4 Pro and NousCoder-14B.
Where can I try DeepSeek-V3.2-Speciale?
On its official website, huggingface.co.
Does DeepSeek-V3.2-Speciale support tool calling?
No. DeepSeek built Speciale only for deep reasoning tasks, and its model card says it does not support tool calling. The standard DeepSeek-V3.2 is the one that calls tools, in both thinking and non-thinking modes.







