DeepSeek R1 Distill Qwen 32B
By DeepSeek
Technical reasoning.
An open-weight model we’re considering for the African-hosted pilot. Explore its publisher documentation and help us understand where it fits your work.
Model capabilities
Text, Reasoning, Code. These describe the publisher’s model, not a tested Embiro service.
Read the official model cardPublished benchmarks
Pass@1 · %. DeepSeek distilled-model evaluation. 64 responses per question; temperature 0.6, top-p 0.95, maximum generation 32,768 tokens.
DeepSeek evaluation · checked 8 October 2026Pass@1 · %. DeepSeek distilled-model evaluation under the same published evaluation settings.
DeepSeek evaluation · checked 8 October 2026Pass@1 · %. DeepSeek distilled-model evaluation. Results describe this reported benchmark version, not a current live leaderboard.
DeepSeek evaluation · checked 8 October 2026Publisher-reported results, not Embiro tests. Scores across different suites or evaluation setups should not be compared.
Before the pilot
We’ll evaluate hardware fit, licensing, quality and serving requirements. Model identifiers, supported features, limits and pricing will be confirmed before access opens.