Model comparison
Claude Sonnet 4 vs DeepSeek-V3.2-Exp
DeepSeek-V3.2-Exp is the stronger model overall, scoring 44.3 to 40.8 on the Noometry Index.
Last verified . 36 shared benchmarks.
DeepSeekSummary
- They share 36 benchmarks with published results for both. Claude Sonnet 4 scores higher in 3 categories and DeepSeek-V3.2-Exp in 6 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in long context, where DeepSeek-V3.2-Exp leads 47.6 to 33.7.
- The biggest single-benchmark swing is Fiction.LiveBench: 46.9% for Claude Sonnet 4 and 83.3% for DeepSeek-V3.2-Exp.
- DeepSeek-V3.2-Exp is cheaper at $0.26 / $0.38 per million input/output tokens, against $3 / $15 for Claude Sonnet 4.
- Claude Sonnet 4 accepts more context: 200K tokens versus 164K.
- DeepSeek-V3.2-Exp has downloadable open weights; the other is API-only.
Side by side
| Claude Sonnet 4 | DeepSeek-V3.2-Exp | |
|---|---|---|
| Provider | Anthropic | DeepSeek |
| Noometry Index | 40.8 | 44.3 |
| Released | 2025-05-22 | 2025-09-29 |
| Weights | Proprietary | Open |
| Context window | 200K | 164K |
| Max output | 64K | 66K |
| Input $ / M tokens | $3 | $0.26 |
| Output $ / M tokens | $15 | $0.38 |
| Results tracked | 58 | 49 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3.2-Exp leads
Claude Sonnet 4: 43.5 (#88), DeepSeek-V3.2-Exp: 46.5 (#65)
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Exp |
|---|---|---|
| SWE-bench Verified (bash only) | 64.9% | 70% |
| Aider Polyglot | 61.3% | 74.2% |
| SciCode | 40% | 38.9% |
| WeirdML | 46.1% | 39.5% |
| LMArena Coding | 1414 | 1454 |
| LMArena WebDev | — | 1362 |
| SWE-bench Multilingual | — | 59% |
| GSO | 4.9% | — |
| ALE-Bench | 655.35 | — |
Agentic & Tool Use Claude Sonnet 4 leads
Claude Sonnet 4: 38.5 (#31), DeepSeek-V3.2-Exp: 32.7 (#59)
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Exp |
|---|---|---|
| TheAgentCompany | 33.1% | 42.9% |
| Terminal-Bench | — | 39.6% |
| APEX-Agents | — | 21.3% |
| Berkeley Function Calling Leaderboard | — | 56.7% |
| Cybench | 35% | — |
| DeepResearch Bench | 46.6% | — |
| OSWorld | 43.9% | — |
| METR Time Horizons | 62% | — |
| Vending-Bench 2 | — | 1,034 |
Reasoning Too close to call
Claude Sonnet 4: 22.9 (#187), DeepSeek-V3.2-Exp: 22.1 (#208)
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Exp |
|---|---|---|
| ARC-AGI-2 | 5.9% | 4% |
| Kagi LLM Benchmark | 73% | 52.2% |
| ARC-AGI-1 | 40% | 57% |
| CritPt | 0.3% | 2.9% |
| LMArena Hard Prompts | 1372 | 1434 |
| DTBench | 77.1% | 87.7% |
| LMCA | 29% | 29.1% |
| Epoch Capabilities Index | 141.69 | 146.27 |
| SimpleBench | 45.5% | — |
| NYT Connections (extended) | — | 36.7% |
| Chess Puzzles | — | 14% |
| EnigmaEval | 3.1% | — |
| Thematic Generalization | — | 65% |
| ForecastBench | 60.2 | — |
Math Claude Sonnet 4 leads
Claude Sonnet 4: 43.3 (#80), DeepSeek-V3.2-Exp: 41.7 (#87)
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Exp |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 71.1% | 87.8% |
| LMArena Math | 1375 | 1435 |
| FrontierMath (Feb 2025 set) | 4.1% | 22.1% |
| FrontierMath Tier 4 (v1) | 0% | 2.1% |
| MathArena Final-Answer Competitions | — | 57.7% |
| ProofBench | — | 8% |
| Omni-MATH | 60.2% | — |
| MATH Level 5 | 84.4% | — |
Knowledge DeepSeek-V3.2-Exp leads
Claude Sonnet 4: 41.8 (#108), DeepSeek-V3.2-Exp: 51.7 (#66)
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Exp |
|---|---|---|
| GPQA Diamond | 79.2% | 83.4% |
| Vectara Hallucination Rate | 10.3% | 5.3% |
| LMArena Expert | 1372 | 1436 |
| Humanity's Last Exam | 7.8% | — |
| MMLU-Pro | 84.3% | — |
| Confabulations | 13.2% | — |
| GPQA (HELM) | 70.6% | — |
Multimodal Not comparable
Claude Sonnet 4: 26.2 (#121), DeepSeek-V3.2-Exp: —
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Vision | 1191 | — |
| GeoBench | 37% | — |
| VPCT | 34% | — |
| MindCube | 44.8% | — |
Multilingual DeepSeek-V3.2-Exp leads
Claude Sonnet 4: 46.7 (#156), DeepSeek-V3.2-Exp: 52.2 (#90)
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Non-English | 1333 | 1409 |
| LMArena Chinese | 1350 | 1461 |
| LMArena French | 1363 | 1433 |
| LMArena German | 1331 | 1440 |
| LMArena Japanese | 1302 | 1374 |
| LMArena Korean | 1291 | 1371 |
| LMArena Russian | 1355 | 1424 |
| LMArena Spanish | 1357 | 1440 |
Instruction Following DeepSeek-V3.2-Exp leads
Claude Sonnet 4: 71.7 (#145), DeepSeek-V3.2-Exp: 74.5 (#93)
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Instruction Following | 1376 | 1413 |
| IFEval | 84% | — |
Long Context DeepSeek-V3.2-Exp leads
Claude Sonnet 4: 33.7 (#259), DeepSeek-V3.2-Exp: 47.6 (#16)
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Exp |
|---|---|---|
| Fiction.LiveBench | 46.9% | 83.3% |
| LMArena Longer Query | 1398 | 1428 |
| CL-bench | — | 13.2% |
| CL-bench Life | — | 9.5% |
Writing & Preference DeepSeek-V3.2-Exp leads
Claude Sonnet 4: 57.1 (#132), DeepSeek-V3.2-Exp: 62.4 (#77)
| Benchmark | Claude Sonnet 4 | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Text | 1351 | 1425 |
| LMArena Creative Writing | 1345 | 1403 |
| EQ-Bench Creative Writing | 1483 | 1515 |
| LMArena Multi-Turn | 1376 | 1427 |
| Short-Story Creative Writing | 81.4% | — |
| WildBench | 83.8% | — |
Frequently asked questions
Is Claude Sonnet 4 better than DeepSeek-V3.2-Exp?
DeepSeek-V3.2-Exp is the stronger model overall, scoring 44.3 to 40.8 on the Noometry Index.
Which is cheaper, Claude Sonnet 4 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; Claude Sonnet 4 lists at $3 and $15.
Is Claude Sonnet 4 or DeepSeek-V3.2-Exp better for coding?
DeepSeek-V3.2-Exp scores higher on coding benchmarks: 46.5 versus 43.5 in the Noometry coding category.
Which has the bigger context window?
Claude Sonnet 4 does, with 200K tokens against 164K.
How many benchmarks do Claude Sonnet 4 and DeepSeek-V3.2-Exp share?
36 benchmarks have published results for both models. Claude Sonnet 4 has 58 scored results on Noometry and DeepSeek-V3.2-Exp has 49.