Model comparison
DeepSeek-R1 vs GPT-4
DeepSeek-R1 is the stronger model overall, scoring 42.3 to 29.1 on the Noometry Index.
Last verified . 25 shared benchmarks.
DeepSeekSummary
- They share 25 benchmarks with published results for both. DeepSeek-R1 scores higher in 8 categories and GPT-4 in 0 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek-R1 leads 43.8 to 10.8.
- The biggest single-benchmark swing is MATH Level 5: 96.6% for DeepSeek-R1 and 23% for GPT-4.
- DeepSeek-R1 is cheaper at $0.50 / $2.15 per million input/output tokens, against $30 / $60 for GPT-4.
- DeepSeek-R1 accepts more context: 164K tokens versus 8K.
Side by side
| DeepSeek-R1 | GPT-4 | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 42.3 | 29.1 |
| Released | 2025-01-20 | 2023-03-14 |
| Weights | Proprietary | Proprietary |
| Context window | 164K | 8K |
| Max output | 64K | 8K |
| Input $ / M tokens | $0.50 | $30 |
| Output $ / M tokens | $2.15 | $60 |
| Results tracked | 52 | 38 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-R1 leads
DeepSeek-R1: 46.3 (#68), GPT-4: 31.6 (#283)
| Benchmark | DeepSeek-R1 | GPT-4 |
|---|---|---|
| WeirdML | 41.6% | 12.4% |
| LMArena Coding | 1427 | 1254 |
| Aider Polyglot | 71.4% | — |
| SciCode | 35.7% | — |
| BigCodeBench Instruct | — | 46% |
| LiveBench Coding | 66.7% | — |
| BigCodeBench Complete | — | 57.2% |
| ALE-Bench | 804.12 | — |
| AlgoTune | 1.7 | — |
| HumanEval+ | — | 79.3% |
Agentic & Tool Use Not comparable
DeepSeek-R1: 30.7 (#75), GPT-4: —
| Benchmark | DeepSeek-R1 | GPT-4 |
|---|---|---|
| METR Time Horizons | 53.8% | 36.1% |
| DeepResearch Bench | 35.1% | — |
| BALROG | 34.9% | — |
Reasoning Too close to call
DeepSeek-R1: 18.6 (#278), GPT-4: 17.8 (#289)
| Benchmark | DeepSeek-R1 | GPT-4 |
|---|---|---|
| LMArena Hard Prompts | 1416 | 1241 |
| Epoch Capabilities Index | 141.29 | 125.89 |
| ForecastBench | 60 | 57.8 |
| ARC-AGI-2 | 1.3% | — |
| SimpleBench | 40.8% | — |
| Kagi LLM Benchmark | 69.4% | — |
| ARC-AGI-1 | 21.2% | — |
| CritPt | 1.1% | — |
| Chess Puzzles | — | 4% |
| LiveBench Reasoning | 83.2% | — |
| Mystery Game Puzzles | — | 12% |
| DTBench | — | 62.7% |
| LiveBench Data Analysis | 69.8% | — |
| LMCA | — | 17.1% |
| BIG-Bench Hard | — | 75.1% |
| HellaSwag | — | 95.3% |
| LiveBench | 71.6% | — |
| WinoGrande | — | 87.5% |
Math DeepSeek-R1 leads
DeepSeek-R1: 43.8 (#79), GPT-4: 10.8 (#309)
| Benchmark | DeepSeek-R1 | GPT-4 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 66.4% | 1.1% |
| LMArena Math | 1400 | 1269 |
| MATH Level 5 | 96.6% | 23% |
| Omni-MATH | 42.4% | — |
| LiveBench Math | 80.7% | — |
| GSM8K | — | 92% |
Knowledge DeepSeek-R1 leads
DeepSeek-R1: 44.5 (#87), GPT-4: 18.4 (#282)
| Benchmark | DeepSeek-R1 | GPT-4 |
|---|---|---|
| GPQA Diamond | 76.3% | 35.7% |
| LMArena Expert | 1394 | 1211 |
| MMLU-Pro | 79.3% | — |
| Confabulations | 12.7% | — |
| Vectara Hallucination Rate | 11.3% | — |
| GPQA (HELM) | 66.6% | — |
| MMLU | — | 86.4% |
| TriviaQA | — | 84.8% |
Multilingual DeepSeek-R1 leads
DeepSeek-R1: 52.4 (#85), GPT-4: 40.6 (#215)
| Benchmark | DeepSeek-R1 | GPT-4 |
|---|---|---|
| LMArena Non-English | 1412 | 1246 |
| LMArena Chinese | 1442 | 1242 |
| LMArena French | 1417 | 1283 |
| LMArena German | 1404 | 1251 |
| LMArena Japanese | 1391 | 1209 |
| LMArena Korean | 1360 | 1184 |
| LMArena Russian | 1423 | 1251 |
| LMArena Spanish | 1411 | 1261 |
Instruction Following DeepSeek-R1 leads
DeepSeek-R1: 72.0 (#143), GPT-4: 65.3 (#222)
| Benchmark | DeepSeek-R1 | GPT-4 |
|---|---|---|
| LMArena Instruction Following | 1382 | 1241 |
| LiveBench Instruction Following | 80.5% | — |
| IFEval | 78.4% | — |
Long Context DeepSeek-R1 leads
DeepSeek-R1: 45.4 (#36), GPT-4: 37.7 (#212)
| Benchmark | DeepSeek-R1 | GPT-4 |
|---|---|---|
| LMArena Longer Query | 1391 | 1244 |
| Fiction.LiveBench | 75% | — |
Writing & Preference DeepSeek-R1 leads
DeepSeek-R1: 61.4 (#88), GPT-4: 34.9 (#268)
| Benchmark | DeepSeek-R1 | GPT-4 |
|---|---|---|
| LMArena Text | 1428 | 1263 |
| LMArena Creative Writing | 1405 | 1244 |
| EQ-Bench Creative Writing | 1500 | 752 |
| LMArena Multi-Turn | 1405 | 1257 |
| Short-Story Creative Writing | 83% | — |
| WildBench | 82.8% | — |
| LiveBench Language | 48.5% | — |
Frequently asked questions
Is DeepSeek-R1 better than GPT-4?
DeepSeek-R1 is the stronger model overall, scoring 42.3 to 29.1 on the Noometry Index.
Which is cheaper, DeepSeek-R1 or GPT-4?
DeepSeek-R1 is cheaper. It lists at $0.50 per million input tokens and $2.15 per million output tokens; GPT-4 lists at $30 and $60.
Is DeepSeek-R1 or GPT-4 better for coding?
DeepSeek-R1 scores higher on coding benchmarks: 46.3 versus 31.6 in the Noometry coding category.
Which has the bigger context window?
DeepSeek-R1 does, with 164K tokens against 8K.
How many benchmarks do DeepSeek-R1 and GPT-4 share?
25 benchmarks have published results for both models. DeepSeek-R1 has 52 scored results on Noometry and GPT-4 has 38.