Short answer
The supplied Ollama listing excerpt says deepseek-v3.2 was retired on July 15, 2026, and that no models had been pushed. That does not establish a current pull-and-run setup or local hardware requirements. DeepSeek’s open-weight announcement links to the V3.2 model, but an open-weight release alone does not confirm Ollama packaging or provide Ollama instructions. The listing and announcement are linked in the sources below.
Can you run V3.2 with Ollama?
The available listing evidence does not provide a model package to pull, a working command, or local setup instructions for deepseek-v3.2. So I can’t recommend ollama run deepseek-v3.2 as a working installation command based on that evidence.
This is not proof that V3.2 can never be run locally. A separately obtained or converted model with custom configuration may be a different route, but the available evidence does not verify a package, quantization, Ollama version, memory requirement, or performance for one. Treat those details as unknown rather than inferring them from another DeepSeek model or an older guide.
V3.2 or V3.2-Speciale?
- DeepSeek-V3.2: DeepSeek described it as balancing inference and response length, and said it supports tool use in both thinking and non-thinking modes. These are the company’s release claims, not local Ollama performance results.
- DeepSeek-V3.2-Speciale: DeepSeek described Speciale as a higher-reasoning variant with greater token use and no tool calls at launch. Its announcement gave a temporary API endpoint with an end date of December 15, 2025, at 15:59 UTC; that date has passed, so the announcement does not establish a currently working endpoint.
BentoML’s model guide positions standard V3.2 for everyday Q&A and general agent tasks, and Speciale as research-oriented rather than optimized for everyday chat or writing. That is the guide’s characterization, not evidence that either variant is currently available in Ollama. (api-docs.deepseek.com)
What if you need an API for an application?
If you specifically require DeepSeek-V3.2, its current API availability is not established by the documentation reviewed here. DeepSeek’s current quick-start and pricing pages surface deepseek-flash and deepseek-v4-pro; they identify deepseek-flash as served by DeepSeek-V4.1-Flash-1017. That may be an option if you can use a currently documented model instead of V3.2, but it is not a way to call V3.2. (api-docs.deepseek.com)
For a basic request to the documented API, DeepSeek’s quick-start shows this OpenAI-compatible chat-completions pattern. Set DEEPSEEK_API_KEY to an API key you obtained from DeepSeek. The example uses the current deepseek-flash model ID; it does not call V3.2. (api-docs.deepseek.com)
curl https://api.deepseek.com/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer ${DEEPSEEK_API_KEY}" \
-d '{
"model": "deepseek-flash",
"messages": [
{"role": "user", "content": "Hello"}
],
"stream": false
}'
The current model-list endpoint is documented as the source for available model identifiers and metadata. Check it and the pricing page before choosing a model or estimating costs; the documented lineup and rates can change. (api-docs.deepseek.com)
Which route should you choose?
| Your requirement | Practical choice |
|---|---|
| You need the exact V3.2 model in Ollama | The supplied Ollama listing excerpt reports the listing retired with no models pushed; a verified pull-and-run setup is not established here. |
| You need DeepSeek through an API and can use a current model | Follow the current official API docs. They document deepseek-flash and deepseek-v4-pro; confirm the live model list, model behavior, and pricing before integrating. |
| You specifically need V3.2 or Speciale | Confirm present availability and an exact model identifier with the provider. Current V3.2 API availability is not established here, and the historical Speciale endpoint’s stated expiry has passed. |
A hosted API means you call a provider’s service rather than setting up local model serving. Local execution is a different deployment arrangement, but the evidence here does not establish a working Ollama V3.2 package. If your application depends on V3.2-specific behavior, verify the exact model and endpoint with the provider before building around it; otherwise, evaluate a currently documented model against your workload.
Sources
DeepSeek-V3.2 Release — DeepSeek API Docs
The Complete Guide to DeepSeek Models: V3, R1, V4 and Beyond — BentoML
Your First API Call — DeepSeek API Docs