DeepSeek takes on o1 with new reasoning preview
DeepSeek just launched DeepSeek-R1-Lite-Preview, delivering OpenAI o1-preview-level performance on AIME and MATH benchmarks. The model lets users watch its thought process unfold in real time, with open-source weights and API access slated to arrive soon.
OpenAI proved with o1 that giving models time to think pays off, but accessible alternatives have been basically non-existent.
Now DeepSeek is stepping up. The team just announced DeepSeek-R1-Lite-Preview, putting supercharged reasoning capabilities directly into users' hands via their web chat.
Showing its work to match o1-preview
According to DeepSeek, the new model achieves performance on par with OpenAI's o1-preview on tough math benchmarks like AIME and MATH.
Instead of making users stare at a vague loading icon, R1-Lite-Preview prints out its reasoning chain in real time. You can actually watch how it breaks down problems step by step.
Right now, anyone can test it for free on DeepSeek's web interface. Even better for builders, the company confirmed that open-source models and an API are coming soon.
Why it matters
For regular users, this is an immediate, free alternative to test frontier-level logic without shelling out for an expensive monthly subscription.
We think the real story here is the promise of open weights. If DeepSeek can actually deliver an open-source reasoning model that rivals o1-preview, it completely changes the economics for developers building complex workflows.
That said, it is still labeled as a "Lite-Preview." How well it handles messy real-world coding or multi-turn logic outside pure math benchmarks remains an open question.
Sources (1)
- @deepseek_ai(opens in a new tab)Originaltwitter.com
Topic trackerDeepSeek-R1 release process
- DeepSeek plans to open-source 5 core repos next week
- DeepSeek drops R1, open-sourcing an o1-level reasoning model under MIT
- DeepSeek takes on o1 with new reasoning previewThis story