DeepSeek-R1: benchmark scores, price and ranking
DeepSeek-R1 from DeepSeek ranks #389 of 743 LLMs on the LLMs Tiger aggregate benchmark score (48.0), based on 13 benchmark results captured up to 2026-10-13.
- Organisation
- DeepSeek
- Type
- Open-weight / local model
- Open weights
- Yes
- Released
- 2025-01-20 (source: curated)
- Context window
- 64,000 tokens
- Cheapest API price
- $0.400 in / $0.400 out per million tokens (via hyperbolic, observed 2026-10-13T19:47Z)
- Aggregate rank
- #389 of 743 (score 48.0)
Relative to other models, DeepSeek-R1 is strongest on MMLU-Pro (#17 of 136) and weakest on ARC-AGI-2 (#68 of 87).
Across the 13 benchmarks where at least ten models are scored, DeepSeek-R1 is in the top 10% on 0 and in the top half on 6.
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 benchmark results
| Benchmark | Score | Rank | Source | Captured |
|---|---|---|---|---|
| MMLU-Pro | 84.0% | #17 of 136 | HuggingFace MMLU-Pro leaderboard | 2026-10-13 |
| Fiction.liveBench | 69.4% | #11 of 47 | Epoch AI Benchmarking Hub | 2026-10-13 |
| BALROG | 34.9% | #9 of 34 | Epoch AI Benchmarking Hub | 2026-10-13 |
| Epoch ECI | 139 | #56 of 204 | Epoch AI Benchmarking Hub | 2026-10-13 |
| Aider Polyglot | 56.9% | #14 of 49 | Aider polyglot leaderboard | 2026-10-13 |
| METR time horizon | 27 min | #15 of 31 | Epoch AI Benchmarking Hub | 2026-10-13 |
| GPQA Diamond | 71.5% | #140 of 275 | HuggingFace GPQA Diamond leaderboard | 2026-10-13 |
| OTIS Mock AIME | 53.3% | #103 of 189 | Epoch AI Benchmarking Hub | 2026-10-13 |
| CritPt | 1.1% | #68 of 115 | Epoch AI Benchmarking Hub | 2026-10-13 |
| SimpleBench | 30.9% | #42 of 61 | Epoch AI Benchmarking Hub | 2026-10-13 |
| WeirdML | 36.5% | #76 of 109 | Epoch AI Benchmarking Hub | 2026-10-13 |
| SciCode | 35.7% | #80 of 106 | Epoch AI Benchmarking Hub | 2026-10-13 |
| ARC-AGI-2 | 1.3% | #68 of 87 | ARC Prize official leaderboard | 2026-10-13 |
About DeepSeek-R1
DeepSeek-R1 is an open-weight model that can run on your own hardware from DeepSeek. It was released on 2025-01-20, according to the developer's announcement. It accepts up to 64,000 tokens of input at once. The cheapest API listing we track charges $0.400 per million input tokens and $0.400 per million output tokens, through hyperbolic. We hold 13 benchmark results for it across 4 areas, and none is in the top tenth of its board.
Where DeepSeek-R1 stands
On our aggregate score, DeepSeek-R1 is #389 of 743 ranked models, around the middle of the table. Its score is 48.0; the scale is centred on 50, so it sits 2.0 points below the typical model. That beats 48% of the ranked field. Among the 33 ranked DeepSeek models it is #19. The next one up is DeepSeek-R1-Distill-Llama-70B at 48.0. The next one down is DeepSeek-R1-Distill-Qwen-32B at 47.9.
DeepSeek-R1, benchmark by benchmark
MMLU-Pro puts DeepSeek-R1 at 84.0%, #17 of 136. That beats 88% of the models tested, but 4.1 points short of Darwin-180B-RSI (88.1%). The results either side are Step-3.5-Flash (84.4%) and K-EXAONE-236B-A23B (83.8%). It covers 57-subject knowledge + reasoning. The board median is 55.3%, so DeepSeek-R1 is 28.7 points above it. Source: HuggingFace MMLU-Pro leaderboard, captured 2026-10-13.
On Fiction.liveBench, DeepSeek-R1 reaches 69.4%. It ranks #11 of 47, ahead of 77% of the field, and 25.0 points behind grok-4-0709 at 94.4%. The results either side are Qwen3-32B (74.2%) and MiniMax-M1-80k (69.4%). It covers long-context story comprehension (16k). The board median is 61.1%, so DeepSeek-R1 is 8.3 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
DeepSeek-R1 manages 34.9% on BALROG, good for #9 of 34. GPT-6 Astra (max) leads at 68.3%, 33.4 points further up, and 74% of models sit below. The results either side are Gemini 2.5 Pro Exp (Mar 2025) (43.3%) and gemini-2.5-flash (33.5%). It covers long-horizon game-playing agents. The board median is 27.6%, so DeepSeek-R1 is 7.3 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
Epoch ECI puts DeepSeek-R1 at 139, #56 of 204. That beats 73% of the models tested, but 17 points short of Qwen 3.8 Max (156). The results either side are qwen3-235b-a22b (139) and Qwen3-235B-A22B-Instruct (Jul 2025) (139). It covers Epoch Capabilities Index: Epoch's own aggregate of ~40 benchmarks. The board median is 127, so DeepSeek-R1 is 12 points above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
On Aider Polyglot, DeepSeek-R1 reaches 56.9%. It ranks #14 of 49, ahead of 71% of the field, and 26.2 points behind gemini-2.5-pro-preview-06-05 at 83.1%. The results either side are Kimi K2 (59.1%) and DeepSeek V3 0324 (55.1%). It covers edits across 6 languages (225 exercises). The board median is 44.9%, so DeepSeek-R1 is 12.0 points above it. Source: Aider polyglot leaderboard, captured 2026-10-13.
DeepSeek-R1 manages 27 min on METR time horizon, good for #15 of 31. Claude Mythos Preview (Early) leads at 1,045 min, 1018 minutes further up, and 52% of models sit below. The results either side are DeepSeek-R1 (May 2025) (31 min) and DeepSeek-V3 (Mar 2025) (23 min). It covers length of task (minutes) done at 50% success. The board median is 23 min, so DeepSeek-R1 is 4 minutes above it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
DeepSeek-R1 lands below the middle on SimpleBench: 30.9%, #42 of 61, with 67% of the field scoring higher. Claude Fable 5 (max) leads at 81.9%; the closest result above is qwen3-235b-a22b at 31.0%. It covers trick questions that test common sense. The board median is 41.4%, so DeepSeek-R1 is 10.5 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
DeepSeek-R1 lands below the middle on WeirdML: 36.5%, #76 of 109, with 69% of the field scoring higher. GPT-6 Astra (max) leads at 93.3%; the closest result above is Kimi-K2-Instruct-0905 at 36.7%. It covers unusual machine-learning tasks. The board median is 41.2%, so DeepSeek-R1 is 4.7 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
DeepSeek-R1 lands below the middle on SciCode: 35.7%, #80 of 106, with 75% of the field scoring higher. Claude Opus 5.5 (max) leads at 66.9%; the closest result above is Qwen3.6 35B-A3B at 35.8%. It covers scientific research code. The board median is 44.3%, so DeepSeek-R1 is 8.6 points below it. Source: Epoch AI Benchmarking Hub, captured 2026-10-13.
DeepSeek-R1 lands below the middle on ARC-AGI-2: 1.3%, #68 of 87, with 77% of the field scoring higher. GPT-6 Astra (max) leads at 95.0%; the closest result above is GPT-5.1 (Thinking, Low) at 1.9%. It covers abstract reasoning, ARC Prize public eval set. The board median is 22.8%, so DeepSeek-R1 is 21.5 points below it. Source: ARC Prize official leaderboard, captured 2026-10-13.
The full table above also lists the other 3 results, including boards with fewer than ten models, where a rank says little.
Strengths and weaknesses
Software engineering: uneven, ahead of 48% of the field on average across 2 benchmarks. Its best board here is Aider Polyglot (#14 of 49) and its weakest is SciCode (#80 of 106). Reasoning, maths and science: uneven, ahead of 46% of the field on average across 6 benchmarks. Its best board here is MMLU-Pro (#17 of 136) and its weakest is ARC-AGI-2 (#68 of 87). General knowledge and chat: solid, ahead of 58% of the field on average across 4 benchmarks. Its best board here is Fiction.liveBench (#11 of 47) and its weakest is WeirdML (#76 of 109).
Price and value
The median input price across the 427 priced models is $0.500 per million tokens, so DeepSeek-R1 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), Qwen2.5-Coder-3B-Instruct ($0.010, 46.5). Similar scores cost more on Kimi K2.5 (instant) ($0.450) and Gemini 3 Flash Preview (High) ($0.500).
Release history
The DeepSeek model released before it was DeepSeek-V3 on 2024-12-26, which scores 45.1. The next one was DeepSeek-R1-Distill-Llama-70B on 2025-01-20, at 48.0.
Reading these numbers
The 13 scores on this page come from 5 sources: Epoch AI Benchmarking Hub (9), ARC Prize official leaderboard (1), Aider polyglot leaderboard (1), HuggingFace GPQA Diamond leaderboard (1). 2 of them come from an official leaderboard and 11 from datasets or third-party runs.
Models ranked nearby
- #386 Gemini 3.5 Flash-Lite (High) (48.0)
- #387 DeepSeek-V3.1 (48.0)
- #388 DeepSeek-R1-Distill-Llama-70B (48.0)
- #390 qwen3-coder-plus (48.0)
- #391 Codex + GPT-6 Luna (48.0)
- #392 Codestral-22B-v0.1-Q4_K_M (48.0)
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