DeepSeek-R1 — reviews, specs & pricing
DeepSeek's open reasoning model rivaling OpenAI o1.
Summary
DeepSeek-R1 is a reasoning-focused model trained with large-scale reinforcement learning, achieving performance comparable to OpenAI's o1 on math and coding benchmarks. It's released fully open-weight under MIT license. It sparked wide adoption due to low cost and openness.
Sample use case
Used for research into reasoning models, math/coding benchmarks, and cost-efficient reasoning applications. Widely used for distillation into smaller models.
Specifications
- Provider: deepseek
- License: open
- Parameters: 671B
- Context: 128k tokens
- Input price: $0.55/M tok
- Output price: $2.19/M tok
- Released: 2025-01-20
Pros
- MIT licensed, fully open
- Strong reasoning performance
- Very low API cost
Cons
- Verbose chain-of-thought output
- Requires significant hosting compute
- Some safety/alignment concerns noted
Average rating 3.6 from 5 community reviews on Reviuws.
Community reviews
Simon Willison's blog on DeepSeek-R1
Rating: 4.0 / 5 — by Simon Willison (use case: local open reasoning models)
Willison calls DeepSeek-R1 a hugely significant open-weights release, noting that it and the distilled Llama and Qwen variants are genuinely usable reasoning models he could run and inspect locally under an MIT licence.
Pros: Open MIT-licensed weights with visible reasoning traces; distilled sizes run on ordinary hardware.
Cons: Confusing naming across the R1 family; distilled models give up some capability.
Ars Technica on DeepSeek-R1
Rating: 4.0 / 5 — by Benj Edwards (use case: self-hosted reasoning)
Ars reports DeepSeek-R1 rivals OpenAI's o1 on several benchmarks while being free to download, run locally and modify, calling it a milestone in Chinese labs catching up with top US reasoning models.
Pros: Matches o1 on several benchmarks at far lower training cost; free open weights.
Cons: Content moderation on sensitive topics; limited independent verification at the time.
WIRED on DeepSeek-R1
Rating: 3.0 / 5 — by Reece Rogers (use case: everyday chatbot use)
WIRED's hands-on against ChatGPT calls DeepSeek a stunning release showing real training-efficiency innovation, but finds hallucinations frequent and answers on politically sensitive China topics restricted.
Pros: Fast, capable responses; efficient training approach.
Cons: Frequent hallucinations; censors politically sensitive questions.
CNET on DeepSeek-R1
Rating: 4.0 / 5 — by Imad Khan (use case: daily reasoning chatbot)
After a week of daily use Khan is highly impressed by R1's visible step-by-step reasoning, while flagging that some topics related to the Chinese government are off limits.
Pros: Transparent, inspectable reasoning; strong for a free model.
Cons: Blocks politically sensitive questions; only a week of testing.
Computerworld on DeepSeek-R1
Rating: 3.0 / 5 — by John E. Dunn (use case: enterprise evaluation)
Written for enterprises, Dunn concludes R1 is genuinely disruptive because it needs far less expensive hardware and undercuts incumbent pricing, while cautioning that cost isn't the whole story.
Pros: Runs on far cheaper hardware; open licensing disrupts platform economics.
Cons: Data governance concerns; total cost and risk picture is more complex.
Frequently asked questions
What is DeepSeek-R1?
DeepSeek-R1 is a reasoning-focused model trained with large-scale reinforcement learning, achieving performance comparable to OpenAI's o1 on math and coding benchmarks. It's released fully open-weight under MIT license. It sparked wide adoption due to low cost and openness.
How much does DeepSeek-R1 cost?
DeepSeek-R1 costs $0.55 per million input tokens and $2.19 per million output tokens.
Is DeepSeek-R1 open source?
DeepSeek-R1 is released under the open license.