Ref https://threadreaderapp.com/thread/1881318130334814301.html
DeepSeek has announced the release of DeepSeek-R1, an open-source AI model comparable in performance to OpenAI's o1. The model is available under the MIT license, allowing for unrestricted use, including distillation and commercialization.
Key highlights include:
Open-Source Distilled Models: Six smaller models distilled from DeepSeek-R1 have been released, with 32B and 70B models performing on par with OpenAI's o1-mini.
License Update: DeepSeek-R1 is now MIT licensed, providing clear open access for the community to utilize model weights and outputs.
Technical Highlights:
Large-scale reinforcement learning applied in post-training.
Significant performance improvements achieved with minimal labeled data.
Enhanced capabilities in mathematics, coding, and reasoning tasks, matching the performance of OpenAI's o1.
API Access & Pricing:
To use DeepSeek-R1, set the model parameter to deepseek-reasoner.
Pricing details:
- $0.14 per million input tokens (cache hit)
- $0.55 per million input tokens (cache miss)
- $2.19 per million output tokens
For more information, visit DeepSeek's officialw ebsite and API documentation.
