Model comparison
DeepSeek-R1 vs GPT-5
GPT-5 is the stronger model overall, scoring 50.9 to 42.3 on the Noometry Index. DeepSeek-R1 costs 3.8× less per token, which makes it the better buy when GPT-5's lead doesn't matter for your workload.
Last verified . 45 shared benchmarks.
DeepSeekSummary
- They share 45 benchmarks with published results for both. DeepSeek-R1 scores higher in 1 category and GPT-5 in 8 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in long context, where GPT-5 leads 69.5 to 45.4.
- The biggest single-benchmark swing is ARC-AGI-1: 21.2% for DeepSeek-R1 and 65.7% for GPT-5.
- DeepSeek-R1 is cheaper at $0.50 / $2.15 per million input/output tokens, against $1.25 / $10 for GPT-5.
- GPT-5 accepts more context: 400K tokens versus 164K.
Side by side
| DeepSeek-R1 | GPT-5 | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 42.3 | 50.9 |
| Released | 2025-01-20 | 2025-08-07 |
| Weights | Proprietary | Proprietary |
| Context window | 164K | 400K |
| Max output | 64K | 128K |
| Input $ / M tokens | $0.50 | $1.25 |
| Output $ / M tokens | $2.15 | $10 |
| Results tracked | 52 | 69 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5 leads
DeepSeek-R1: 46.3 (#68), GPT-5: 50.3 (#47)
| Benchmark | DeepSeek-R1 | GPT-5 |
|---|---|---|
| Aider Polyglot | 71.4% | 88% |
| SciCode | 35.7% | 42.9% |
| WeirdML | 41.6% | 60.7% |
| LMArena Coding | 1427 | 1436 |
| ALE-Bench | 804.12 | 1,162 |
| AlgoTune | 1.7 | 1.67 |
| SWE-bench Verified | — | 73.6% |
| SWE-bench Verified (bash only) | — | 65% |
| LMArena WebDev | — | 1418 |
| GSO | — | 6.9% |
| LiveBench Coding | 66.7% | — |
Agentic & Tool Use GPT-5 leads
DeepSeek-R1: 30.7 (#75), GPT-5: 33.1 (#56)
| Benchmark | DeepSeek-R1 | GPT-5 |
|---|---|---|
| DeepResearch Bench | 35.1% | 49.6% |
| BALROG | 34.9% | 32.8% |
| METR Time Horizons | 53.8% | 69.6% |
| Terminal-Bench | — | 49.6% |
| GDPval | — | 34.8% |
| Remote Labor Index | — | 1.7% |
| LMArena Search | — | 1133 |
Reasoning GPT-5 leads
DeepSeek-R1: 18.6 (#278), GPT-5: 38.3 (#64)
| Benchmark | DeepSeek-R1 | GPT-5 |
|---|---|---|
| ARC-AGI-2 | 1.3% | 9.9% |
| SimpleBench | 40.8% | 56.7% |
| Kagi LLM Benchmark | 69.4% | 72.7% |
| ARC-AGI-1 | 21.2% | 65.7% |
| CritPt | 1.1% | 12.6% |
| LMArena Hard Prompts | 1416 | 1416 |
| Epoch Capabilities Index | 141.29 | 150 |
| ForecastBench | 60 | 61.4 |
| Chess Puzzles | — | 37% |
| EnigmaEval | — | 10.5% |
| EBR-Bench | — | 12.7% |
| LiveBench Reasoning | 83.2% | — |
| Mystery Game Puzzles | — | 23% |
| DTBench | — | 90.7% |
| LiveBench Data Analysis | 69.8% | — |
| LMCA | — | 40% |
| LiveBench | 71.6% | — |
Math GPT-5 leads
DeepSeek-R1: 43.8 (#79), GPT-5: 55.0 (#44)
| Benchmark | DeepSeek-R1 | GPT-5 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 66.4% | 91.4% |
| Omni-MATH | 42.4% | 64.7% |
| LMArena Math | 1400 | 1407 |
| MATH Level 5 | 96.6% | 98.1% |
| FrontierMath (Tiers 1-3) | — | 55.4% |
| FrontierMath Tier 4 | — | 22% |
| ProofBench | — | 18% |
| LiveBench Math | 80.7% | — |
| FrontierMath (Feb 2025 set) | — | 32.4% |
| FrontierMath Tier 4 (v1) | — | 12.5% |
Knowledge GPT-5 leads
DeepSeek-R1: 44.5 (#87), GPT-5: 56.6 (#43)
| Benchmark | DeepSeek-R1 | GPT-5 |
|---|---|---|
| GPQA Diamond | 76.3% | 86.2% |
| MMLU-Pro | 79.3% | 86.3% |
| Confabulations | 12.7% | 10.3% |
| Vectara Hallucination Rate | 11.3% | 14.7% |
| GPQA (HELM) | 66.6% | 79.2% |
| LMArena Expert | 1394 | 1419 |
| Humanity's Last Exam | — | 25.3% |
| SimpleQA Verified | — | 50.1% |
Multimodal Not comparable
DeepSeek-R1: —, GPT-5: 46.8 (#13)
| Benchmark | DeepSeek-R1 | GPT-5 |
|---|---|---|
| LMArena Vision | — | 1232 |
| GeoBench | — | 81% |
| VPCT | — | 66% |
Multilingual DeepSeek-R1 leads
DeepSeek-R1: 52.4 (#85), GPT-5: 51.4 (#110)
| Benchmark | DeepSeek-R1 | GPT-5 |
|---|---|---|
| LMArena Non-English | 1412 | 1397 |
| LMArena Chinese | 1442 | 1422 |
| LMArena French | 1417 | 1410 |
| LMArena German | 1404 | 1416 |
| LMArena Japanese | 1391 | 1409 |
| LMArena Korean | 1360 | 1360 |
| LMArena Russian | 1423 | 1406 |
| LMArena Spanish | 1411 | 1399 |
Instruction Following GPT-5 leads
DeepSeek-R1: 72.0 (#143), GPT-5: 73.8 (#113)
| Benchmark | DeepSeek-R1 | GPT-5 |
|---|---|---|
| IFEval | 78.4% | 87.5% |
| LMArena Instruction Following | 1382 | 1388 |
| LiveBench Instruction Following | 80.5% | — |
Long Context GPT-5 leads
DeepSeek-R1: 45.4 (#36), GPT-5: 69.5 (#2)
| Benchmark | DeepSeek-R1 | GPT-5 |
|---|---|---|
| Fiction.LiveBench | 75% | 97.2% |
| LMArena Longer Query | 1391 | 1399 |
Writing & Preference GPT-5 leads
DeepSeek-R1: 61.4 (#88), GPT-5: 63.4 (#65)
| Benchmark | DeepSeek-R1 | GPT-5 |
|---|---|---|
| LMArena Text | 1428 | 1406 |
| LMArena Creative Writing | 1405 | 1365 |
| Short-Story Creative Writing | 83% | 86% |
| EQ-Bench Creative Writing | 1500 | 1627 |
| WildBench | 82.8% | 85.7% |
| LMArena Multi-Turn | 1405 | 1426 |
| LiveBench Language | 48.5% | — |
Frequently asked questions
Is DeepSeek-R1 better than GPT-5?
GPT-5 is the stronger model overall, scoring 50.9 to 42.3 on the Noometry Index. DeepSeek-R1 costs 3.8× less per token, which makes it the better buy when GPT-5's lead doesn't matter for your workload.
Which is cheaper, DeepSeek-R1 or GPT-5?
DeepSeek-R1 is cheaper. It lists at $0.50 per million input tokens and $2.15 per million output tokens; GPT-5 lists at $1.25 and $10.
Is DeepSeek-R1 or GPT-5 better for coding?
GPT-5 scores higher on coding benchmarks: 50.3 versus 46.3 in the Noometry coding category.
Which has the bigger context window?
GPT-5 does, with 400K tokens against 164K.
How many benchmarks do DeepSeek-R1 and GPT-5 share?
45 benchmarks have published results for both models. DeepSeek-R1 has 52 scored results on Noometry and GPT-5 has 69.