Model comparison
DeepSeek-V3.2-Exp vs o3
o3 is the stronger model overall, scoring 47.5 to 44.3 on the Noometry Index. DeepSeek-V3.2-Exp costs 12× less per token, which makes it the better buy when o3's lead doesn't matter for your workload.
Last verified . 36 shared benchmarks.
DeepSeekSummary
- They share 36 benchmarks with published results for both. DeepSeek-V3.2-Exp scores higher in 2 categories and o3 in 7 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where o3 leads 32.0 to 22.1.
- The biggest single-benchmark swing is Chess Puzzles: 14% for DeepSeek-V3.2-Exp and 38% for o3.
- DeepSeek-V3.2-Exp is cheaper at $0.26 / $0.38 per million input/output tokens, against $2 / $8 for o3.
- o3 accepts more context: 200K tokens versus 164K.
- DeepSeek-V3.2-Exp has downloadable open weights; the other is API-only.
Side by side
| DeepSeek-V3.2-Exp | o3 | |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 44.3 | 47.5 |
| Released | 2025-09-29 | 2025-04-16 |
| Weights | Open | Proprietary |
| Context window | 164K | 200K |
| Max output | 66K | 100K |
| Input $ / M tokens | $0.26 | $2 |
| Output $ / M tokens | $0.38 | $8 |
| Results tracked | 49 | 63 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
DeepSeek-V3.2-Exp: 46.5 (#65), o3: 46.8 (#64)
| Benchmark | DeepSeek-V3.2-Exp | o3 |
|---|---|---|
| SWE-bench Verified (bash only) | 70% | 58.4% |
| Aider Polyglot | 74.2% | 81.3% |
| WeirdML | 39.5% | 52.4% |
| LMArena Coding | 1454 | 1408 |
| SWE-bench Verified | — | 62.3% |
| LMArena WebDev | 1362 | — |
| SWE-bench Multilingual | 59% | — |
| SciCode | 38.9% | — |
| GSO | — | 8.8% |
| CadEval | — | 74% |
| ALE-Bench | — | 933.55 |
Agentic & Tool Use o3 leads
DeepSeek-V3.2-Exp: 32.7 (#59), o3: 34.5 (#44)
| Benchmark | DeepSeek-V3.2-Exp | o3 |
|---|---|---|
| Berkeley Function Calling Leaderboard | 56.7% | 63% |
| Terminal-Bench | 39.6% | — |
| APEX-Agents | 21.3% | — |
| GDPval | — | 30.8% |
| TheAgentCompany | 42.9% | — |
| DeepResearch Bench | — | 45.2% |
| OSWorld | — | 23% |
| LMArena Search | — | 1144 |
| METR Time Horizons | — | 65.4% |
| Vending-Bench 2 | 1,034 | — |
Reasoning o3 leads
DeepSeek-V3.2-Exp: 22.1 (#208), o3: 32.0 (#78)
| Benchmark | DeepSeek-V3.2-Exp | o3 |
|---|---|---|
| ARC-AGI-2 | 4% | 6.5% |
| Kagi LLM Benchmark | 52.2% | 67.6% |
| ARC-AGI-1 | 57% | 60.8% |
| CritPt | 2.9% | 1.4% |
| Chess Puzzles | 14% | 38% |
| LMArena Hard Prompts | 1434 | 1402 |
| DTBench | 87.7% | 84.8% |
| LMCA | 29.1% | 39.7% |
| Epoch Capabilities Index | 146.27 | 146.86 |
| SimpleBench | — | 53.1% |
| NYT Connections (extended) | 36.7% | — |
| EnigmaEval | — | 13.1% |
| Thematic Generalization | 65% | — |
| Mystery Game Puzzles | — | 29% |
| ForecastBench | — | 62.5 |
Math o3 leads
DeepSeek-V3.2-Exp: 41.7 (#87), o3: 50.2 (#58)
| Benchmark | DeepSeek-V3.2-Exp | o3 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 87.8% | 84.4% |
| LMArena Math | 1435 | 1426 |
| FrontierMath (Feb 2025 set) | 22.1% | 18.7% |
| FrontierMath Tier 4 (v1) | 2.1% | 2.1% |
| FrontierMath (Tiers 1-3) | — | 33.3% |
| MathArena Final-Answer Competitions | 57.7% | — |
| ProofBench | 8% | — |
| Omni-MATH | — | 71.4% |
| MATH Level 5 | — | 97.8% |
Knowledge o3 leads
DeepSeek-V3.2-Exp: 51.7 (#66), o3: 54.6 (#52)
| Benchmark | DeepSeek-V3.2-Exp | o3 |
|---|---|---|
| GPQA Diamond | 83.4% | 81.8% |
| LMArena Expert | 1436 | 1402 |
| Humanity's Last Exam | — | 20.3% |
| SimpleQA Verified | — | 49.4% |
| MMLU-Pro | — | 85.9% |
| Confabulations | — | 14.4% |
| Vectara Hallucination Rate | 5.3% | — |
| GPQA (HELM) | — | 75.3% |
Multimodal Not comparable
DeepSeek-V3.2-Exp: —, o3: 41.4 (#36)
| Benchmark | DeepSeek-V3.2-Exp | o3 |
|---|---|---|
| LMArena Vision | — | 1214 |
| GeoBench | — | 74% |
| VPCT | — | 52% |
Multilingual Too close to call
DeepSeek-V3.2-Exp: 52.2 (#90), o3: 51.7 (#105)
| Benchmark | DeepSeek-V3.2-Exp | o3 |
|---|---|---|
| LMArena Non-English | 1409 | 1401 |
| LMArena Chinese | 1461 | 1437 |
| LMArena French | 1433 | 1430 |
| LMArena German | 1440 | 1420 |
| LMArena Japanese | 1374 | 1403 |
| LMArena Korean | 1371 | 1370 |
| LMArena Russian | 1424 | 1406 |
| LMArena Spanish | 1440 | 1395 |
Instruction Following DeepSeek-V3.2-Exp leads
DeepSeek-V3.2-Exp: 74.5 (#93), o3: 72.8 (#127)
| Benchmark | DeepSeek-V3.2-Exp | o3 |
|---|---|---|
| LMArena Instruction Following | 1413 | 1368 |
| IFEval | — | 86.9% |
Long Context o3 leads
DeepSeek-V3.2-Exp: 47.6 (#16), o3: 53.3 (#6)
| Benchmark | DeepSeek-V3.2-Exp | o3 |
|---|---|---|
| Fiction.LiveBench | 83.3% | 88.9% |
| CL-bench | 13.2% | 17.8% |
| LMArena Longer Query | 1428 | 1372 |
| CL-bench Life | 9.5% | — |
Writing & Preference o3 leads
DeepSeek-V3.2-Exp: 62.4 (#77), o3: 63.5 (#64)
| Benchmark | DeepSeek-V3.2-Exp | o3 |
|---|---|---|
| LMArena Text | 1425 | 1410 |
| LMArena Creative Writing | 1403 | 1359 |
| EQ-Bench Creative Writing | 1515 | 1676 |
| LMArena Multi-Turn | 1427 | 1405 |
| Short-Story Creative Writing | — | 83.9% |
| WildBench | — | 86.1% |
Frequently asked questions
Is DeepSeek-V3.2-Exp better than o3?
o3 is the stronger model overall, scoring 47.5 to 44.3 on the Noometry Index. DeepSeek-V3.2-Exp costs 12× less per token, which makes it the better buy when o3's lead doesn't matter for your workload.
Which is cheaper, DeepSeek-V3.2-Exp or o3?
DeepSeek-V3.2-Exp is cheaper. It lists at $0.26 per million input tokens and $0.38 per million output tokens; o3 lists at $2 and $8.
Is DeepSeek-V3.2-Exp or o3 better for coding?
They score almost the same on coding (46.5 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-V3.2-Exp and o3 share?
36 benchmarks have published results for both models. DeepSeek-V3.2-Exp has 49 scored results on Noometry and o3 has 63.