DeepSeek model product
DeepSeek-R1
DeepSeek-R1 is a reasoning-focused language model from DeepSeek that features advanced thinking capabilities.
Updated Aug 12, 2026. Default version: DeepSeek-R1-0528
LLMBoard score44.2DeepSeek-R1-0528
Coverage80%16 benchmark families
Context window131.1KTokens
Official input priceN/AOfficial price unavailable
DeepSeek-R1 Specifications
Technical details for the model's default version.
- Version
- DeepSeek-R1-0528
- Released
- May 28, 2025
- Knowledge cutoff
- Unknown
- Parameters
- 671B
- Context window
- 131.1K
- Max output
- 131.1K
- Inputs
- text
- Outputs
- text
- Open weights
- No
- License
- MIT
DeepSeek-R1 Capability Profile
This profile uses the model's current scored version. Arena ratings and prices are shown separately.
DeepSeek-R1 Benchmark Results
Benchmark scores for DeepSeek-R1-0528.
| SimpleQA | 92.3% | 04 | 46 | 93.3% | C | |
| Aider-Polyglot | 71.6% | 06 | 22 | 76.2% | C | |
| AIME 2024 | 91.4% | 07 | 53 | 88.5% | C | |
| MMLU-Redux | 93.4% | 10 | 48 | 80.8% | C | |
| BrowseComp-zh | 35.7% | 13 | 13 | 0.0% | C | |
| CodeForces | 64.3% | 14 | 16 | 13.3% | C | |
| LiveCodeBench | 73.3% | 17 | 73 | 77.8% | C | |
| MMLU-Pro | 85.0% | 17 | 129 | 87.5% | C | |
| HMMT 2025 | 79.4% | 25 | 33 | 25.0% | C | |
| Terminal-Bench | 5.7% | 25 | 25 | 0.0% | C | |
| SWE-bench Multilingual | 30.5% | 34 | 34 | 0.0% | C | |
| AIME 2025 | 87.5% | 55 | 114 | 52.2% | C | |
| BrowseComp | 8.9% | 58 | 58 | 0.0% | C | |
| Humanity's Last Exam | 17.7% | 64 | 93 | 31.5% | C | |
| GPQA | 81.0% | 81 | 234 | 65.7% | C | |
| SWE-Bench Verified | 44.6% | 92 | 105 | 12.5% | C |
DeepSeek-R1 Arena Results
Preference and agent-evaluation results for the default version.
| text | overall | 80 | 1427.8 | 18,498 | N/A | |
| text style control | overall | 107 | 1421.9 | 18,498 | N/A |
DeepSeek-R1 Runtime Performance
Provider-specific output speed and catalog latency for DeepSeek-R1-0528. Runtime does not affect the capability score.
| DeepInfra | 45.04 tok/s | 0.61 s | 131.1K | 131.1K | |
| Novita | 37.96 tok/s | 1.18 s | 131.1K | 32.8K | |
| DeepSeek | 9 tok/s | 0.3 s | 131.1K | 131.1K |
Output Speed is generated output tokens received per second. Catalog latency is reported separately from observed provider TTFT.
DeepSeek-R1 Pricing
Official vendor API pricing appears first, followed by individual provider offers.
- Official API
- N/A
- Official provider
- N/A
- Lowest third-party
- From $0.40 input, $1.7 output per 1M via NanoGPT
- Tracked offerings
- 12
| NanoGPT | deepseek-ai/DeepSeek-R1-0528 | global | $0.40 | $1.7 | 128K | |
| submodel | deepseek-ai/DeepSeek-R1-0528 | global | $0.50 | $2.15 | 75K | |
| Deep Infra | deepseek-ai/DeepSeek-R1-0528 | global | $0.50 | $2.15 | 163.8K | |
| Meganova | deepseek-ai/DeepSeek-R1-0528 | global | $0.50 | $2.15 | 163.8K | |
| Alibaba (China) | deepseek-r1-0528 | global | $0.574 | $2.29 | 131.1K | |
| Cortecs | deepseek-r1-0528 | global | $0.652 | $2.57 | 164K | |
| TensorX | deepseek/deepseek-r1-0528 | global | $0.66 | $2.6 | 164K | |
| Kilo Gateway | deepseek/deepseek-r1-0528 | global | $0.70 | $2.5 | 163.8K | |
| NovitaAI | deepseek/deepseek-r1-0528 | global | $0.70 | $2.5 | 163.8K | |
| Jiekou.AI | deepseek/deepseek-r1-0528 | global | $0.70 | $2.5 | 163.8K | |
| Hugging Face | deepseek-ai/DeepSeek-R1-0528 | global | $3 | $5 | 163.8K |
Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.
DeepSeek-R1 Versions
Available versions of this model. The score column identifies the version used in the overall ranking.
| DeepSeek-R1-0528 | 44.2 | 671B | 131.1K | 131.1K | No | MIT | |
| DeepSeek-R1 | N/A | 671B | 128K | 32.8K | Yes | MIT | |
| DeepSeek R1 Zero | N/A | 671B | N/A | N/A | No | MIT |
What is DeepSeek-R1?
Key information about DeepSeek-R1 and its available data.
DeepSeek-R1 is a reasoning-focused language model from DeepSeek that features advanced thinking capabilities. It serves as the foundation for DeepSeek's reasoning model family and pioneered their thinking mode approach for complex problem-solving tasks.
Data as of 2026-08-11.
DeepSeek-R1 vs nearby models
Open a comparison with the three ranked models immediately above and below this model.
Models similar to DeepSeek-R1
Recommendations prioritize the same model type and family, then the closest LLMBoard score.
DE
DeepSeek-V3.1
DeepSeek
DE
DeepSeek-V3.2
DeepSeek
DE
DeepSeek-V3
DeepSeek
DE
DeepSeek-R1-Distill-Qwen
DeepSeek
DE
DeepSeek-V2.5
DeepSeek
DE
DeepSeek-V4-Pro
DeepSeek
FAQ
Common questions about DeepSeek-R1.
When was DeepSeek-R1 released?
DeepSeek-R1's default version was released on May 28, 2025.
How much does DeepSeek-R1 cost?
No official standard PAYG price is currently available for DeepSeek-R1. The lowest tracked third-party offer starts at $0.40 input and $1.7 output via NanoGPT.
Who created DeepSeek-R1?
DeepSeek-R1 was created by DeepSeek.
What is the context window for DeepSeek-R1?
The default version has a 131.1K token context window.
Is DeepSeek-R1 open weight?
No. The default version is not marked as having publicly available weights.
How many API providers offer DeepSeek-R1?
12 provider offerings are linked to the default version.
What models should I compare DeepSeek-R1 with?
Nearby ranked alternatives include K EXAONE 236B A23B, Nemotron 3 Super, Sarvam 105B.