Model comparison
DeepSeek-V3 vs DeepSeek-V3.2-Exp
DeepSeek-V3.2-Exp is the stronger model overall, scoring 44.9 to 39.2 on the Noometry Index.
Last verified . 31 shared benchmarks.
DeepSeekSummary
- They share 31 benchmarks with published results for both. DeepSeek-V3 scores higher in 0 categories and DeepSeek-V3.2-Exp in 8 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in knowledge, where DeepSeek-V3.2-Exp leads 51.6 to 37.5.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 37.8% for DeepSeek-V3 and 87.8% for DeepSeek-V3.2-Exp.
- DeepSeek-V3.2-Exp is cheaper at $0.26 / $0.38 per million input/output tokens, against $0.24 / $0.90 for DeepSeek-V3.
Side by side
| DeepSeek-V3 | DeepSeek-V3.2-Exp | |
|---|---|---|
| Provider | DeepSeek | DeepSeek |
| Noometry Index | 39.2 | 44.9 |
| Released | 2024-12-26 | 2025-09-29 |
| Weights | Open | Open |
| Context window | 164K | 164K |
| Max output | 164K | 66K |
| Input $ / M tokens | $0.24 | $0.26 |
| Output $ / M tokens | $0.90 | $0.38 |
| Results tracked | 60 | 48 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3.2-Exp leads
DeepSeek-V3: 41.9 (#99), DeepSeek-V3.2-Exp: 46.4 (#61)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.2-Exp |
|---|---|---|
| Aider Polyglot | 55.1% | 74.2% |
| SciCode | 35.8% | 38.9% |
| WeirdML | 36.1% | 39.5% |
| LMArena Coding | 1368 | 1454 |
| SWE-bench Verified (bash only) | — | 70% |
| LMArena WebDev | — | 1362 |
| SWE-bench Multilingual | — | 59% |
| BigCodeBench Instruct | 50% | — |
| LiveBench Coding | 70.9% | — |
| BigCodeBench Complete | 62.2% | — |
| HumanEval+ | 86.6% | — |
| MBPP+ | 73% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, DeepSeek-V3.2-Exp: 37.9 (#31)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.2-Exp |
|---|---|---|
| Terminal-Bench | — | 39.6% |
| Berkeley Function Calling Leaderboard | — | 56.7% |
| TheAgentCompany | — | 42.9% |
| METR Time Horizons | 49.6% | — |
| Vending-Bench 2 | — | 1,034 |
Reasoning DeepSeek-V3.2-Exp leads
DeepSeek-V3: 20.0 (#226), DeepSeek-V3.2-Exp: 22.0 (#185)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.2-Exp |
|---|---|---|
| Kagi LLM Benchmark | 52.3% | 52.2% |
| CritPt | 0% | 2.9% |
| LMArena Hard Prompts | 1365 | 1433 |
| DTBench | 64.8% | 87.7% |
| LMCA | 15.5% | 29.1% |
| Epoch Capabilities Index | 135.95 | 146.28 |
| ARC-AGI-2 | — | 4% |
| SimpleBench | 27.2% | — |
| NYT Connections (extended) | — | 36.7% |
| ARC-AGI-1 | — | 57% |
| Chess Puzzles | — | 14% |
| Thematic Generalization | — | 65% |
| LiveBench Reasoning | 65.8% | — |
| LiveBench Data Analysis | 60.9% | — |
| BIG-Bench Hard | 87.5% | — |
| ForecastBench | 59.1 | — |
| HellaSwag | 88.9% | — |
| LiveBench | 66.9% | — |
| PIQA | 84.7% | — |
| WinoGrande | 85.2% | — |
Math DeepSeek-V3.2-Exp leads
DeepSeek-V3: 30.4 (#208), DeepSeek-V3.2-Exp: 40.8 (#75)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.2-Exp |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 37.8% | 87.8% |
| LMArena Math | 1373 | 1435 |
| FrontierMath (Feb 2025 set) | 1.7% | 22.1% |
| MathArena Final-Answer Competitions | — | 57.7% |
| ProofBench | — | 8% |
| Omni-MATH | 40.3% | — |
| LiveBench Math | 73.5% | — |
| MATH Level 5 | 75.5% | — |
| FrontierMath Tier 4 (v1) | — | 2.1% |
Knowledge DeepSeek-V3.2-Exp leads
DeepSeek-V3: 37.5 (#147), DeepSeek-V3.2-Exp: 51.6 (#61)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.2-Exp |
|---|---|---|
| GPQA Diamond | 67.6% | 83.4% |
| Vectara Hallucination Rate | 6.1% | 5.3% |
| LMArena Expert | 1352 | 1436 |
| MMLU-Pro | 72.3% | — |
| Confabulations | 26.1% | — |
| GPQA (HELM) | 53.8% | — |
| ARC (AI2) Challenge | 95.3% | — |
| MMLU | 87.2% | — |
| TriviaQA | 82.9% | — |
Multilingual DeepSeek-V3.2-Exp leads
DeepSeek-V3: 48.8 (#138), DeepSeek-V3.2-Exp: 52.4 (#85)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Non-English | 1358 | 1409 |
| LMArena Chinese | 1390 | 1461 |
| LMArena French | 1386 | 1435 |
| LMArena German | 1374 | 1440 |
| LMArena Japanese | 1333 | 1377 |
| LMArena Korean | 1321 | 1372 |
| LMArena Russian | 1372 | 1423 |
| LMArena Spanish | 1358 | 1441 |
Instruction Following DeepSeek-V3.2-Exp leads
DeepSeek-V3: 73.0 (#125), DeepSeek-V3.2-Exp: 74.7 (#89)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Instruction Following | 1345 | 1413 |
| LiveBench Instruction Following | 81.5% | — |
| IFEval | 83.2% | — |
Long Context DeepSeek-V3.2-Exp leads
DeepSeek-V3: 34.0 (#248), DeepSeek-V3.2-Exp: 47.6 (#15)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.2-Exp |
|---|---|---|
| Fiction.LiveBench | 50% | 83.3% |
| LMArena Longer Query | 1353 | 1428 |
| CL-bench | — | 13.2% |
| CL-bench Life | — | 9.5% |
Writing & Preference DeepSeek-V3.2-Exp leads
DeepSeek-V3: 57.6 (#124), DeepSeek-V3.2-Exp: 62.5 (#73)
| Benchmark | DeepSeek-V3 | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Text | 1375 | 1425 |
| LMArena Creative Writing | 1364 | 1403 |
| EQ-Bench Creative Writing | 1472 | 1515 |
| LMArena Multi-Turn | 1389 | 1427 |
| Short-Story Creative Writing | 77% | — |
| WildBench | 83% | — |
| LiveBench Language | 49.1% | — |
Frequently asked questions
Is DeepSeek-V3 better than DeepSeek-V3.2-Exp?
DeepSeek-V3.2-Exp is the stronger model overall, scoring 44.9 to 39.2 on the Noometry Index.
Which is cheaper, DeepSeek-V3 or DeepSeek-V3.2-Exp?
DeepSeek-V3.2-Exp is cheaper. It lists at $0.26 per million input tokens and $0.38 per million output tokens; DeepSeek-V3 lists at $0.24 and $0.90.
Is DeepSeek-V3 or DeepSeek-V3.2-Exp better for coding?
DeepSeek-V3.2-Exp scores higher on coding benchmarks: 46.4 versus 41.9 in the Noometry coding category.
Which has the bigger context window?
Both accept 164K tokens.
How many benchmarks do DeepSeek-V3 and DeepSeek-V3.2-Exp share?
31 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and DeepSeek-V3.2-Exp has 48.