Model comparison
DeepSeek V4 Pro vs o3
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 47.5 on the Noometry Index.
Last verified . 35 shared benchmarks.
DeepSeekSummary
- They share 35 benchmarks with published results for both. DeepSeek V4 Pro scores higher in 7 categories and o3 in 2 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek V4 Pro leads 56.5 to 32.0.
- The biggest single-benchmark swing is ARC-AGI-2: 61.3% for DeepSeek V4 Pro and 6.5% for o3.
- DeepSeek V4 Pro is cheaper at $0.66 / $1.98 per million input/output tokens, against $2 / $8 for o3.
- DeepSeek V4 Pro accepts more context: 1M tokens versus 200K.
- DeepSeek V4 Pro has downloadable open weights; the other is API-only.
Side by side
| DeepSeek V4 Pro | o3 | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 54.3 | 47.5 |
| Released | 2026-04-24 | 2025-04-16 |
| Weights | Open | Proprietary |
| Context window | 1M | 200K |
| Max output | 393K | 100K |
| Input $ / M tokens | $0.66 | $2 |
| Output $ / M tokens | $1.98 | $8 |
| Results tracked | 48 | 63 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek V4 Pro leads
DeepSeek V4 Pro: 52.4 (#34), o3: 46.8 (#64)
| Benchmark | DeepSeek V4 Pro | o3 |
|---|---|---|
| SWE-bench Verified | 77.6% | 62.3% |
| WeirdML | 66.2% | 52.4% |
| LMArena Coding | 1470 | 1408 |
| ALE-Bench | 1,403 | 933.55 |
| FrontierCode | 28.6% | — |
| SWE-bench Verified (bash only) | — | 58.4% |
| Aider Polyglot | — | 81.3% |
| LMArena WebDev | 1582 | — |
| SciCode | 51% | — |
| GSO | — | 8.8% |
| CadEval | — | 74% |
Agentic & Tool Use o3 leads
DeepSeek V4 Pro: 32.8 (#58), o3: 34.5 (#44)
| Benchmark | DeepSeek V4 Pro | o3 |
|---|---|---|
| APEX-Agents | 47.3% | — |
| Berkeley Function Calling Leaderboard | — | 63% |
| GDPval | — | 30.8% |
| DeepResearch Bench | — | 45.2% |
| OSWorld | — | 23% |
| LMArena Search | — | 1144 |
| METR Time Horizons | — | 65.4% |
| Vending-Bench 2 | 3,285 | — |
Reasoning DeepSeek V4 Pro leads
DeepSeek V4 Pro: 56.5 (#24), o3: 32.0 (#78)
| Benchmark | DeepSeek V4 Pro | o3 |
|---|---|---|
| ARC-AGI-2 | 61.3% | 6.5% |
| Kagi LLM Benchmark | 53.5% | 67.6% |
| ARC-AGI-1 | 90.5% | 60.8% |
| CritPt | 18% | 1.4% |
| Chess Puzzles | 47% | 38% |
| LMArena Hard Prompts | 1461 | 1402 |
| Mystery Game Puzzles | 43% | 29% |
| DTBench | 93.9% | 84.8% |
| LMCA | 45.5% | 39.7% |
| Epoch Capabilities Index | 155.31 | 146.86 |
| ForecastBench | 56.1 | 62.5 |
| SimpleBench | — | 53.1% |
| NYT Connections (extended) | 91.3% | — |
| EnigmaEval | — | 13.1% |
| Surface Evolver Bench | 40% | — |
Math DeepSeek V4 Pro leads
DeepSeek V4 Pro: 64.8 (#30), o3: 50.2 (#58)
| Benchmark | DeepSeek V4 Pro | o3 |
|---|---|---|
| FrontierMath (Tiers 1-3) | 64.6% | 33.3% |
| OTIS Mock AIME 2024-2025 | 98.6% | 84.4% |
| LMArena Math | 1455 | 1426 |
| FrontierMath Tier 4 | 26.8% | — |
| MathArena Final-Answer Competitions | 76.6% | — |
| ProofBench | 50% | — |
| Omni-MATH | — | 71.4% |
| MATH Level 5 | — | 97.8% |
| FrontierMath (Feb 2025 set) | — | 18.7% |
| FrontierMath Tier 4 (v1) | — | 2.1% |
Knowledge DeepSeek V4 Pro leads
DeepSeek V4 Pro: 59.5 (#31), o3: 54.6 (#52)
| Benchmark | DeepSeek V4 Pro | o3 |
|---|---|---|
| GPQA Diamond | 91.7% | 81.8% |
| SimpleQA Verified | 52.9% | 49.4% |
| LMArena Expert | 1464 | 1402 |
| Humanity's Last Exam | — | 20.3% |
| MMLU-Pro | — | 85.9% |
| Confabulations | — | 14.4% |
| Vectara Hallucination Rate | 8.6% | — |
| GPQA (HELM) | — | 75.3% |
Multimodal Not comparable
DeepSeek V4 Pro: —, o3: 41.4 (#36)
| Benchmark | DeepSeek V4 Pro | o3 |
|---|---|---|
| LMArena Vision | — | 1214 |
| GeoBench | — | 74% |
| VPCT | — | 52% |
Multilingual DeepSeek V4 Pro leads
DeepSeek V4 Pro: 54.4 (#45), o3: 51.7 (#105)
| Benchmark | DeepSeek V4 Pro | o3 |
|---|---|---|
| LMArena Non-English | 1439 | 1401 |
| LMArena Chinese | 1486 | 1437 |
| LMArena French | 1472 | 1430 |
| LMArena German | 1458 | 1420 |
| LMArena Japanese | 1445 | 1403 |
| LMArena Korean | 1447 | 1370 |
| LMArena Russian | 1453 | 1406 |
| LMArena Spanish | 1458 | 1395 |
Instruction Following DeepSeek V4 Pro leads
DeepSeek V4 Pro: 76.1 (#47), o3: 72.8 (#127)
| Benchmark | DeepSeek V4 Pro | o3 |
|---|---|---|
| LMArena Instruction Following | 1448 | 1368 |
| IFEval | — | 86.9% |
Long Context o3 leads
DeepSeek V4 Pro: 45.0 (#51), o3: 53.3 (#6)
| Benchmark | DeepSeek V4 Pro | o3 |
|---|---|---|
| LMArena Longer Query | 1458 | 1372 |
| Fiction.LiveBench | — | 88.9% |
| CL-bench | — | 17.8% |
| CL-bench Life | 13.5% | — |
Writing & Preference DeepSeek V4 Pro leads
DeepSeek V4 Pro: 65.5 (#46), o3: 63.5 (#64)
| Benchmark | DeepSeek V4 Pro | o3 |
|---|---|---|
| LMArena Text | 1451 | 1410 |
| LMArena Creative Writing | 1446 | 1359 |
| EQ-Bench Creative Writing | 1553 | 1676 |
| LMArena Multi-Turn | 1467 | 1405 |
| Short-Story Creative Writing | — | 83.9% |
| WildBench | — | 86.1% |
| EQ-Bench 4 | 1166 | — |
Frequently asked questions
Is DeepSeek V4 Pro better than o3?
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 47.5 on the Noometry Index.
Which is cheaper, DeepSeek V4 Pro or o3?
DeepSeek V4 Pro is cheaper. It lists at $0.66 per million input tokens and $1.98 per million output tokens; o3 lists at $2 and $8.
Is DeepSeek V4 Pro or o3 better for coding?
DeepSeek V4 Pro scores higher on coding benchmarks: 52.4 versus 46.8 in the Noometry coding category.
Which has the bigger context window?
DeepSeek V4 Pro does, with 1M tokens against 200K.
How many benchmarks do DeepSeek V4 Pro and o3 share?
35 benchmarks have published results for both models. DeepSeek V4 Pro has 48 scored results on Noometry and o3 has 63.