DeepSeek-R1 (May 2025): benchmark scores, price and ranking
DeepSeek-R1 (May 2025) from DeepSeek ranks #373 of 743 LLMs on the LLMs Tiger aggregate benchmark score (48.2), based on 9 benchmark results captured up to 2026-10-13.
- Organisation
- DeepSeek
- Type
- Open-weight / local model
- Open weights
- Yes
- Released
- 2025-05-28 (source: epoch)
- Cheapest API price
- $0.400 in / $0.400 out per million tokens (via hyperbolic, observed 2026-10-13T19:47Z)
- Aggregate rank
- #373 of 743 (score 48.2)
Relative to other models, DeepSeek-R1 (May 2025) is strongest on Fiction.liveBench (#8 of 47) and weakest on SimpleBench (#33 of 61).
Across the 8 benchmarks where at least ten models are scored, DeepSeek-R1 (May 2025) is in the top 10% on 0 and in the top half on 7.
Its cheapest listed input price, $0.400 per million tokens, is lower than that of 53% of the 427 models with a listed price.
DeepSeek-R1 (May 2025) benchmark results
| Benchmark | Score | Rank | Source | Captured |
|---|---|---|---|---|
| Fiction.liveBench | 75.0% | #8 of 47 | Epoch AI Benchmarking Hub | 2026-10-13 |
| Epoch ECI | 141 | #47 of 204 | Epoch AI Benchmarking Hub | 2026-10-13 |
| ALE-Bench | 804 | #34 of 79 | Epoch AI Benchmarking Hub | 2026-10-13 |
| GPQA Diamond | 76.3% | #120 of 275 | Epoch AI Benchmarking Hub | 2026-10-13 |
| METR time horizon | 31 min | #14 of 31 | Epoch AI Benchmarking Hub | 2026-10-13 |
| OTIS Mock AIME | 66.4% | #86 of 189 | Epoch AI Benchmarking Hub | 2026-10-13 |
| WeirdML | 41.6% | #54 of 109 | Epoch AI Benchmarking Hub | 2026-10-13 |
| SimpleBench | 40.8% | #33 of 61 | Epoch AI Benchmarking Hub | 2026-10-13 |
| DeepResearch Bench | 35.1% | #7 of 7 | Epoch AI Benchmarking Hub | 2026-10-13 |
About DeepSeek-R1 (May 2025)
DeepSeek-R1 (May 2025) is an open-weight model that can run on your own hardware from DeepSeek. It was released on 2025-05-28, according to Epoch AI. The cheapest API listing we track charges $0.400 per million input tokens and $0.400 per million output tokens, through hyperbolic. We hold 9 benchmark results for it across 4 areas, and none is in the top tenth of its board.
Where DeepSeek-R1 (May 2025) stands
On our aggregate score, DeepSeek-R1 (May 2025) is #373 of 743 ranked models, around the middle of the table. Its score is 48.2; the scale is centred on 50, so it sits 1.8 points below the typical model. That beats 50% of the ranked field. Among the 33 ranked DeepSeek models it is #16. The next one up is DeepSeek V4.1 Flash (high) at 48.9. The next one down is DeepSeek-V3.1 at 48.0.
DeepSeek-R1 (May 2025), benchmark by benchmark
Fiction.liveBench puts DeepSeek-R1 (May 2025) at 75.0%, #8 of 47. That beats 83% of the models tested, but 19.4 points short of grok-4-0709 (94.4%). The results either side are gemini-2.5-flash-preview-05-20 (77.8%) and Qwen3-235B-A22B-Thinking-2507 (75.0%). It covers long-context story comprehension (16k). The board median is 61.1%, so DeepSeek-R1 (May 2025) is 13.9 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
On Epoch ECI, DeepSeek-R1 (May 2025) reaches 141. It ranks #47 of 204, ahead of 77% of the field, and 15 points behind Qwen 3.8 Max at 156. The results either side are Mistral Medium 3.5 (141) and Gemini 2.5 Flash (Jun 2025) (141). It covers Epoch Capabilities Index: Epoch's own aggregate of ~40 benchmarks. The board median is 127, so DeepSeek-R1 (May 2025) is 14 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
DeepSeek-R1 (May 2025) manages 804 on ALE-Bench, good for #34 of 79. GPT-6 Astra (max) leads at 2,951, 2,147 points further up, and 57% of models sit below. The results either side are Kimi K2.5 (822) and GPT-5 mini (high) (800). It covers AtCoder heuristic-optimisation contests (performance rating). The board median is 764, so DeepSeek-R1 (May 2025) is 40 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
GPQA Diamond puts DeepSeek-R1 (May 2025) at 76.3%, #120 of 275. That beats 56% of the models tested, but 19.4 points short of GPT-6 Astra (max) (95.8%). The results either side are o1 (high) (76.8%) and Claude Opus 4 (76.3%). It covers expert-level science questions. The board median is 71.7%, so DeepSeek-R1 (May 2025) is 4.6 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
On METR time horizon, DeepSeek-R1 (May 2025) reaches 31 min. It ranks #14 of 31, ahead of 55% of the field, and 1014 minutes behind Claude Mythos Preview (Early) at 1,045 min. The results either side are Gemini 2.5 Pro Preview (Jun 2025) (39 min) and DeepSeek-R1 (27 min). It covers length of task (minutes) done at 50% success. The board median is 23 min, so DeepSeek-R1 (May 2025) is 8 minutes above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
DeepSeek-R1 (May 2025) manages 66.4% on OTIS Mock AIME, good for #86 of 189. Claude Fable 5.1 (max) leads at 100.0%, 33.6 points further up, and 54% of models sit below. The results either side are claude-haiku-4-5-20251001_32K (66.7%) and qwen3-14b (66.4%). It covers competition maths (AIME 2024-25 style). The board median is 57.8%, so DeepSeek-R1 (May 2025) is 8.6 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
WeirdML puts DeepSeek-R1 (May 2025) at 41.6%, #54 of 109. That beats 50% of the models tested, but 51.6 points short of GPT-6 Astra (max) (93.3%). The results either side are gemini-2.5-flash-preview-09-2025_16k (41.9%) and Qwen3-Coder-480B-A35B-Instruct (41.2%). It covers unusual machine-learning tasks. The board median is 41.2%, so DeepSeek-R1 (May 2025) is 0.5 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
DeepSeek-R1 (May 2025) lands below the middle on SimpleBench: 40.8%, #33 of 61, with 52% of the field scoring higher. Claude Fable 5 (max) leads at 81.9%; the closest result above is gemini-2.5-flash at 41.2%. It covers trick questions that test common sense. The board median is 41.4%, so DeepSeek-R1 (May 2025) is 0.6 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
The full table above also lists the other 1 result, including boards with fewer than ten models, where a rank says little.
Strengths and weaknesses
Reasoning, maths and science: solid, ahead of 52% of the field on average across 3 benchmarks. Its best board here is GPQA Diamond (#120 of 275) and its weakest is SimpleBench (#33 of 61). General knowledge and chat: solid, ahead of 66% of the field on average across 4 benchmarks. Its best board here is Fiction.liveBench (#8 of 47) and its weakest is WeirdML (#54 of 109).
Price and value
The median input price across the 427 priced models is $0.500 per million tokens, so DeepSeek-R1 (May 2025) costs less than the typical listing. Models that score within 1.5 points or better and cost less: Qwen2.5-Coder-7B-Instruct ($0.010, 49.5), DeepSeek V4.1 Flash (high) ($0.010, 48.9), Llama-3.2-3B-Instruct ($0.020, 47.2). Similar scores cost more on mimo-v2.5-pro ($0.435) and Inkling Small (xhigh) ($0.450).
Release history
The DeepSeek model released before it was deepseek-r1-0528-qwen3-8b on 2025-05-28, which scores 43.6. The next one was DeepSeek-V3.1 on 2025-08-21, at 48.0.
Reading these numbers
The 9 scores on this page come from 1 source: Epoch AI Benchmarking Hub (9). No official leaderboard supplied any of them, so treat them as third-party measurements.
Models ranked nearby
- #370 ernie-5.1 (48.3)
- #371 Opus 4.5 (Thinking, 8K) (48.2)
- #372 qwen3.5-35b-a3b (48.2)
- #374 Qwen2.5 (7B + 72B) (48.2)
- #375 claude-3-7-sonnet-20250219 (48.2)
- #376 qwen2:72b-instruct-q8_0 (48.2)
More from DeepSeek
- DeepSeek V4 Pro 0813 (max) (#23, 54.3)
- DeepSeek V4 Flash 0731 (max) (#42, 53.1)
- DeepSeek-V3.2-Exp (high) (#67, 52.4)
- deepseek-v4.1-flash-max (#75, 52.2)
- DeepSeek-V2.5-1210 (#101, 51.7)
- DeepSeek v4 Pro (max) (#108, 51.6)
Compare it on the live leaderboard → · How the aggregate score works