Model comparison
DeepSeek-V3 vs o3
o3 is the stronger model overall, scoring 47.5 to 39.5 on the Noometry Index. DeepSeek-V3 costs 8.6× less per token, which makes it the better buy when o3's lead doesn't matter for your workload.
Last verified . 40 shared benchmarks.
DeepSeekSummary
- They share 40 benchmarks with published results for both. DeepSeek-V3 scores higher in 0 categories and o3 in 8 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in long context, where o3 leads 53.3 to 34.0.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 37.8% for DeepSeek-V3 and 84.4% for o3.
- DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $2 / $8 for o3.
- o3 accepts more context: 200K tokens versus 164K.
- DeepSeek-V3 has downloadable open weights; the other is API-only.
Side by side
| DeepSeek-V3 | o3 | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 39.5 | 47.5 |
| Released | 2024-12-26 | 2025-04-16 |
| Weights | Open | Proprietary |
| Context window | 164K | 200K |
| Max output | 164K | 100K |
| Input $ / M tokens | $0.24 | $2 |
| Output $ / M tokens | $0.90 | $8 |
| Results tracked | 60 | 63 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding o3 leads
DeepSeek-V3: 42.3 (#106), o3: 46.8 (#64)
| Benchmark | DeepSeek-V3 | o3 |
|---|---|---|
| Aider Polyglot | 55.1% | 81.3% |
| WeirdML | 36.1% | 52.4% |
| LMArena Coding | 1368 | 1408 |
| SWE-bench Verified | — | 62.3% |
| SWE-bench Verified (bash only) | — | 58.4% |
| SciCode | 35.8% | — |
| GSO | — | 8.8% |
| BigCodeBench Instruct | 50% | — |
| LiveBench Coding | 70.9% | — |
| BigCodeBench Complete | 62.2% | — |
| CadEval | — | 74% |
| ALE-Bench | — | 933.55 |
| HumanEval+ | 86.6% | — |
| MBPP+ | 73% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, o3: 34.5 (#44)
| Benchmark | DeepSeek-V3 | o3 |
|---|---|---|
| METR Time Horizons | 49.6% | 65.4% |
| Berkeley Function Calling Leaderboard | — | 63% |
| GDPval | — | 30.8% |
| DeepResearch Bench | — | 45.2% |
| OSWorld | — | 23% |
| LMArena Search | — | 1144 |
Reasoning o3 leads
DeepSeek-V3: 20.5 (#236), o3: 32.0 (#78)
| Benchmark | DeepSeek-V3 | o3 |
|---|---|---|
| SimpleBench | 27.2% | 53.1% |
| Kagi LLM Benchmark | 52.3% | 67.6% |
| CritPt | 0% | 1.4% |
| LMArena Hard Prompts | 1365 | 1402 |
| DTBench | 64.8% | 84.8% |
| LMCA | 15.5% | 39.7% |
| Epoch Capabilities Index | 135.94 | 146.86 |
| ForecastBench | 59.1 | 62.5 |
| ARC-AGI-2 | — | 6.5% |
| ARC-AGI-1 | — | 60.8% |
| Chess Puzzles | — | 38% |
| EnigmaEval | — | 13.1% |
| LiveBench Reasoning | 65.8% | — |
| Mystery Game Puzzles | — | 29% |
| LiveBench Data Analysis | 60.9% | — |
| BIG-Bench Hard | 87.5% | — |
| HellaSwag | 88.9% | — |
| LiveBench | 66.9% | — |
| PIQA | 84.7% | — |
| WinoGrande | 85.2% | — |
Math o3 leads
DeepSeek-V3: 32.1 (#219), o3: 50.2 (#58)
| Benchmark | DeepSeek-V3 | o3 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 37.8% | 84.4% |
| Omni-MATH | 40.3% | 71.4% |
| LMArena Math | 1373 | 1426 |
| MATH Level 5 | 75.5% | 97.8% |
| FrontierMath (Feb 2025 set) | 1.7% | 18.7% |
| FrontierMath (Tiers 1-3) | — | 33.3% |
| LiveBench Math | 73.5% | — |
| FrontierMath Tier 4 (v1) | — | 2.1% |
Knowledge o3 leads
DeepSeek-V3: 37.5 (#155), o3: 54.6 (#52)
| Benchmark | DeepSeek-V3 | o3 |
|---|---|---|
| GPQA Diamond | 67.6% | 81.8% |
| MMLU-Pro | 72.3% | 85.9% |
| Confabulations | 26.1% | 14.4% |
| GPQA (HELM) | 53.8% | 75.3% |
| LMArena Expert | 1351 | 1402 |
| Humanity's Last Exam | — | 20.3% |
| SimpleQA Verified | — | 49.4% |
| Vectara Hallucination Rate | 6.1% | — |
| ARC (AI2) Challenge | 95.3% | — |
| MMLU | 87.2% | — |
| TriviaQA | 82.9% | — |
Multimodal Not comparable
DeepSeek-V3: —, o3: 41.4 (#36)
| Benchmark | DeepSeek-V3 | o3 |
|---|---|---|
| LMArena Vision | — | 1214 |
| GeoBench | — | 74% |
| VPCT | — | 52% |
Multilingual o3 leads
DeepSeek-V3: 48.5 (#143), o3: 51.7 (#105)
| Benchmark | DeepSeek-V3 | o3 |
|---|---|---|
| LMArena Non-English | 1358 | 1401 |
| LMArena Chinese | 1391 | 1437 |
| LMArena French | 1385 | 1430 |
| LMArena German | 1374 | 1420 |
| LMArena Japanese | 1333 | 1403 |
| LMArena Korean | 1319 | 1370 |
| LMArena Russian | 1373 | 1406 |
| LMArena Spanish | 1358 | 1395 |
Instruction Following Too close to call
DeepSeek-V3: 72.8 (#130), o3: 72.8 (#127)
| Benchmark | DeepSeek-V3 | o3 |
|---|---|---|
| IFEval | 83.2% | 86.9% |
| LMArena Instruction Following | 1345 | 1368 |
| LiveBench Instruction Following | 81.5% | — |
Long Context o3 leads
DeepSeek-V3: 34.0 (#253), o3: 53.3 (#6)
| Benchmark | DeepSeek-V3 | o3 |
|---|---|---|
| Fiction.LiveBench | 50% | 88.9% |
| LMArena Longer Query | 1352 | 1372 |
| CL-bench | — | 17.8% |
Writing & Preference o3 leads
DeepSeek-V3: 57.4 (#130), o3: 63.5 (#64)
| Benchmark | DeepSeek-V3 | o3 |
|---|---|---|
| LMArena Text | 1375 | 1410 |
| LMArena Creative Writing | 1364 | 1359 |
| Short-Story Creative Writing | 77% | 83.9% |
| EQ-Bench Creative Writing | 1472 | 1676 |
| WildBench | 83% | 86.1% |
| LMArena Multi-Turn | 1389 | 1405 |
| LiveBench Language | 49.1% | — |
Frequently asked questions
Is DeepSeek-V3 better than o3?
o3 is the stronger model overall, scoring 47.5 to 39.5 on the Noometry Index. DeepSeek-V3 costs 8.6× less per token, which makes it the better buy when o3's lead doesn't matter for your workload.
Which is cheaper, DeepSeek-V3 or o3?
DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; o3 lists at $2 and $8.
Is DeepSeek-V3 or o3 better for coding?
o3 scores higher on coding benchmarks: 46.8 versus 42.3 in the Noometry coding category.
Which has the bigger context window?
o3 does, with 200K tokens against 164K.
How many benchmarks do DeepSeek-V3 and o3 share?
40 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and o3 has 63.