Model comparison
DeepSeek-R1 vs o3
o3 is the stronger model overall, scoring 47.5 to 42.3 on the Noometry Index. DeepSeek-R1 costs 3.8× less per token, which makes it the better buy when o3's lead doesn't matter for your workload.
Last verified . 41 shared benchmarks.
DeepSeekSummary
- They share 41 benchmarks with published results for both. DeepSeek-R1 scores higher in 1 category and o3 in 8 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where o3 leads 32.0 to 18.6.
- The biggest single-benchmark swing is ARC-AGI-1: 21.2% for DeepSeek-R1 and 60.8% for o3.
- DeepSeek-R1 is cheaper at $0.50 / $2.15 per million input/output tokens, against $2 / $8 for o3.
- o3 accepts more context: 200K tokens versus 164K.
Side by side
| DeepSeek-R1 | o3 | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 42.3 | 47.5 |
| Released | 2025-01-20 | 2025-04-16 |
| Weights | Proprietary | Proprietary |
| Context window | 164K | 200K |
| Max output | 64K | 100K |
| Input $ / M tokens | $0.50 | $2 |
| Output $ / M tokens | $2.15 | $8 |
| Results tracked | 52 | 63 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
DeepSeek-R1: 46.3 (#68), o3: 46.8 (#64)
| Benchmark | DeepSeek-R1 | o3 |
|---|---|---|
| Aider Polyglot | 71.4% | 81.3% |
| WeirdML | 41.6% | 52.4% |
| LMArena Coding | 1427 | 1408 |
| ALE-Bench | 804.12 | 933.55 |
| SWE-bench Verified | — | 62.3% |
| SWE-bench Verified (bash only) | — | 58.4% |
| SciCode | 35.7% | — |
| GSO | — | 8.8% |
| LiveBench Coding | 66.7% | — |
| CadEval | — | 74% |
| AlgoTune | 1.7 | — |
Agentic & Tool Use o3 leads
DeepSeek-R1: 30.7 (#75), o3: 34.5 (#44)
| Benchmark | DeepSeek-R1 | o3 |
|---|---|---|
| DeepResearch Bench | 35.1% | 45.2% |
| METR Time Horizons | 53.8% | 65.4% |
| Berkeley Function Calling Leaderboard | — | 63% |
| GDPval | — | 30.8% |
| OSWorld | — | 23% |
| BALROG | 34.9% | — |
| LMArena Search | — | 1144 |
Reasoning o3 leads
DeepSeek-R1: 18.6 (#278), o3: 32.0 (#78)
| Benchmark | DeepSeek-R1 | o3 |
|---|---|---|
| ARC-AGI-2 | 1.3% | 6.5% |
| SimpleBench | 40.8% | 53.1% |
| Kagi LLM Benchmark | 69.4% | 67.6% |
| ARC-AGI-1 | 21.2% | 60.8% |
| CritPt | 1.1% | 1.4% |
| LMArena Hard Prompts | 1416 | 1402 |
| Epoch Capabilities Index | 141.29 | 146.86 |
| ForecastBench | 60 | 62.5 |
| Chess Puzzles | — | 38% |
| EnigmaEval | — | 13.1% |
| LiveBench Reasoning | 83.2% | — |
| Mystery Game Puzzles | — | 29% |
| DTBench | — | 84.8% |
| LiveBench Data Analysis | 69.8% | — |
| LMCA | — | 39.7% |
| LiveBench | 71.6% | — |
Math o3 leads
DeepSeek-R1: 43.8 (#79), o3: 50.2 (#58)
| Benchmark | DeepSeek-R1 | o3 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 66.4% | 84.4% |
| Omni-MATH | 42.4% | 71.4% |
| LMArena Math | 1400 | 1426 |
| MATH Level 5 | 96.6% | 97.8% |
| FrontierMath (Tiers 1-3) | — | 33.3% |
| LiveBench Math | 80.7% | — |
| FrontierMath (Feb 2025 set) | — | 18.7% |
| FrontierMath Tier 4 (v1) | — | 2.1% |
Knowledge o3 leads
DeepSeek-R1: 44.5 (#87), o3: 54.6 (#52)
| Benchmark | DeepSeek-R1 | o3 |
|---|---|---|
| GPQA Diamond | 76.3% | 81.8% |
| MMLU-Pro | 79.3% | 85.9% |
| Confabulations | 12.7% | 14.4% |
| GPQA (HELM) | 66.6% | 75.3% |
| LMArena Expert | 1394 | 1402 |
| Humanity's Last Exam | — | 20.3% |
| SimpleQA Verified | — | 49.4% |
| Vectara Hallucination Rate | 11.3% | — |
Multimodal Not comparable
DeepSeek-R1: —, o3: 41.4 (#36)
| Benchmark | DeepSeek-R1 | o3 |
|---|---|---|
| LMArena Vision | — | 1214 |
| GeoBench | — | 74% |
| VPCT | — | 52% |
Multilingual Too close to call
DeepSeek-R1: 52.4 (#85), o3: 51.7 (#105)
| Benchmark | DeepSeek-R1 | o3 |
|---|---|---|
| LMArena Non-English | 1412 | 1401 |
| LMArena Chinese | 1442 | 1437 |
| LMArena French | 1417 | 1430 |
| LMArena German | 1404 | 1420 |
| LMArena Japanese | 1391 | 1403 |
| LMArena Korean | 1360 | 1370 |
| LMArena Russian | 1423 | 1406 |
| LMArena Spanish | 1411 | 1395 |
Instruction Following Too close to call
DeepSeek-R1: 72.0 (#143), o3: 72.8 (#127)
| Benchmark | DeepSeek-R1 | o3 |
|---|---|---|
| IFEval | 78.4% | 86.9% |
| LMArena Instruction Following | 1382 | 1368 |
| LiveBench Instruction Following | 80.5% | — |
Long Context o3 leads
DeepSeek-R1: 45.4 (#36), o3: 53.3 (#6)
| Benchmark | DeepSeek-R1 | o3 |
|---|---|---|
| Fiction.LiveBench | 75% | 88.9% |
| LMArena Longer Query | 1391 | 1372 |
| CL-bench | — | 17.8% |
Writing & Preference o3 leads
DeepSeek-R1: 61.4 (#88), o3: 63.5 (#64)
| Benchmark | DeepSeek-R1 | o3 |
|---|---|---|
| LMArena Text | 1428 | 1410 |
| LMArena Creative Writing | 1405 | 1359 |
| Short-Story Creative Writing | 83% | 83.9% |
| EQ-Bench Creative Writing | 1500 | 1676 |
| WildBench | 82.8% | 86.1% |
| LMArena Multi-Turn | 1405 | 1405 |
| LiveBench Language | 48.5% | — |
Frequently asked questions
Is DeepSeek-R1 better than o3?
o3 is the stronger model overall, scoring 47.5 to 42.3 on the Noometry Index. DeepSeek-R1 costs 3.8× less per token, which makes it the better buy when o3's lead doesn't matter for your workload.
Which is cheaper, DeepSeek-R1 or o3?
DeepSeek-R1 is cheaper. It lists at $0.50 per million input tokens and $2.15 per million output tokens; o3 lists at $2 and $8.
Is DeepSeek-R1 or o3 better for coding?
They score almost the same on coding (46.3 vs 46.8); test both on your own repository before choosing.
Which has the bigger context window?
o3 does, with 200K tokens against 164K.
How many benchmarks do DeepSeek-R1 and o3 share?
41 benchmarks have published results for both models. DeepSeek-R1 has 52 scored results on Noometry and o3 has 63.