Model comparison
Claude Haiku 4.5 vs DeepSeek-V3.2-Exp
DeepSeek-V3.2-Exp is the stronger model overall, scoring 44.9 to 39.3 on the Noometry Index.
Last verified . 38 shared benchmarks.
DeepSeekSummary
- They share 38 benchmarks with published results for both. Claude Haiku 4.5 scores higher in 1 category and DeepSeek-V3.2-Exp in 8 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in knowledge, where DeepSeek-V3.2-Exp leads 51.6 to 37.6.
- The biggest single-benchmark swing is NYT Connections (extended): 14.3% for Claude Haiku 4.5 and 36.7% for DeepSeek-V3.2-Exp.
- DeepSeek-V3.2-Exp is cheaper at $0.26 / $0.38 per million input/output tokens, against $1 / $5 for Claude Haiku 4.5.
- Claude Haiku 4.5 accepts more context: 200K tokens versus 164K.
- DeepSeek-V3.2-Exp has downloadable open weights; the other is API-only.
Side by side
| Claude Haiku 4.5 | DeepSeek-V3.2-Exp | |
|---|---|---|
| Provider | Anthropic | DeepSeek |
| Noometry Index | 39.3 | 44.9 |
| Released | 2025-10-15 | 2025-09-29 |
| Weights | Proprietary | Open |
| Context window | 200K | 164K |
| Max output | 64K | 66K |
| Input $ / M tokens | $1 | $0.26 |
| Output $ / M tokens | $5 | $0.38 |
| Results tracked | 53 | 48 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3.2-Exp leads
Claude Haiku 4.5: 43.9 (#71), DeepSeek-V3.2-Exp: 46.4 (#61)
| Benchmark | Claude Haiku 4.5 | DeepSeek-V3.2-Exp |
|---|---|---|
| SWE-bench Verified (bash only) | 66.6% | 70% |
| LMArena WebDev | 1330 | 1362 |
| SWE-bench Multilingual | 64.7% | 59% |
| SciCode | 43.3% | 38.9% |
| WeirdML | 45.4% | 39.5% |
| LMArena Coding | 1454 | 1454 |
| Aider Polyglot | — | 74.2% |
| ALE-Bench | 653.48 | — |
Agentic & Tool Use DeepSeek-V3.2-Exp leads
Claude Haiku 4.5: 33.5 (#50), DeepSeek-V3.2-Exp: 37.9 (#31)
| Benchmark | Claude Haiku 4.5 | DeepSeek-V3.2-Exp |
|---|---|---|
| Terminal-Bench | 35.5% | 39.6% |
| Berkeley Function Calling Leaderboard | 68.7% | 56.7% |
| Vending-Bench 2 | 458.89 | 1,034 |
| TheAgentCompany | — | 42.9% |
| DeepResearch Bench | 45.5% | — |
| BALROG | 31.2% | — |
| ExploitBench | 13.7% | — |
Reasoning DeepSeek-V3.2-Exp leads
Claude Haiku 4.5: 15.1 (#310), DeepSeek-V3.2-Exp: 22.0 (#185)
| Benchmark | Claude Haiku 4.5 | DeepSeek-V3.2-Exp |
|---|---|---|
| ARC-AGI-2 | 4% | 4% |
| NYT Connections (extended) | 14.3% | 36.7% |
| ARC-AGI-1 | 47.7% | 57% |
| CritPt | 0% | 2.9% |
| Chess Puzzles | 8% | 14% |
| LMArena Hard Prompts | 1420 | 1433 |
| DTBench | 73.6% | 87.7% |
| LMCA | 30.9% | 29.1% |
| Epoch Capabilities Index | 142.42 | 146.28 |
| Kagi LLM Benchmark | — | 52.2% |
| Thematic Generalization | — | 65% |
| ForecastBench | 61.4 | — |
Math Claude Haiku 4.5 leads
Claude Haiku 4.5: 43.0 (#69), DeepSeek-V3.2-Exp: 40.8 (#75)
| Benchmark | Claude Haiku 4.5 | DeepSeek-V3.2-Exp |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 66.7% | 87.8% |
| LMArena Math | 1395 | 1435 |
| FrontierMath (Feb 2025 set) | 5.9% | 22.1% |
| FrontierMath Tier 4 (v1) | 2.1% | 2.1% |
| MathArena Final-Answer Competitions | — | 57.7% |
| ProofBench | — | 8% |
| Omni-MATH | 56.1% | — |
| MATH Level 5 | 96.4% | — |
Knowledge DeepSeek-V3.2-Exp leads
Claude Haiku 4.5: 37.6 (#146), DeepSeek-V3.2-Exp: 51.6 (#61)
| Benchmark | Claude Haiku 4.5 | DeepSeek-V3.2-Exp |
|---|---|---|
| GPQA Diamond | 71.2% | 83.4% |
| Vectara Hallucination Rate | 9.8% | 5.3% |
| LMArena Expert | 1443 | 1436 |
| SimpleQA Verified | 13.2% | — |
| MMLU-Pro | 77.7% | — |
| GPQA (HELM) | 60.5% | — |
Multimodal Not comparable
Claude Haiku 4.5: 27.0 (#111), DeepSeek-V3.2-Exp: —
| Benchmark | Claude Haiku 4.5 | DeepSeek-V3.2-Exp |
|---|---|---|
| Blueprint-Bench 2 | 0% | — |
| LMArena Document | 1420 | — |
Multilingual DeepSeek-V3.2-Exp leads
Claude Haiku 4.5: 50.1 (#124), DeepSeek-V3.2-Exp: 52.4 (#85)
| Benchmark | Claude Haiku 4.5 | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Non-English | 1376 | 1409 |
| LMArena Chinese | 1415 | 1461 |
| LMArena French | 1408 | 1435 |
| LMArena German | 1377 | 1440 |
| LMArena Japanese | 1339 | 1377 |
| LMArena Korean | 1346 | 1372 |
| LMArena Russian | 1380 | 1423 |
| LMArena Spanish | 1420 | 1441 |
Instruction Following DeepSeek-V3.2-Exp leads
Claude Haiku 4.5: 71.6 (#144), DeepSeek-V3.2-Exp: 74.7 (#89)
| Benchmark | Claude Haiku 4.5 | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Instruction Following | 1414 | 1413 |
| IFEval | 80.1% | — |
Long Context DeepSeek-V3.2-Exp leads
Claude Haiku 4.5: 43.6 (#86), DeepSeek-V3.2-Exp: 47.6 (#15)
| Benchmark | Claude Haiku 4.5 | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Longer Query | 1427 | 1428 |
| Fiction.LiveBench | — | 83.3% |
| CL-bench | — | 13.2% |
| CL-bench Life | — | 9.5% |
Writing & Preference DeepSeek-V3.2-Exp leads
Claude Haiku 4.5: 58.1 (#119), DeepSeek-V3.2-Exp: 62.5 (#73)
| Benchmark | Claude Haiku 4.5 | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Text | 1396 | 1425 |
| LMArena Creative Writing | 1371 | 1403 |
| LMArena Multi-Turn | 1409 | 1427 |
| EQ-Bench Creative Writing | — | 1515 |
| WildBench | 83.9% | — |
| EQ-Bench 4 | 1064 | — |
Frequently asked questions
Is Claude Haiku 4.5 better than DeepSeek-V3.2-Exp?
DeepSeek-V3.2-Exp is the stronger model overall, scoring 44.9 to 39.3 on the Noometry Index.
Which is cheaper, Claude Haiku 4.5 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 Haiku 4.5 lists at $1 and $5.
Is Claude Haiku 4.5 or DeepSeek-V3.2-Exp better for coding?
DeepSeek-V3.2-Exp scores higher on coding benchmarks: 46.4 versus 43.9 in the Noometry coding category.
Which has the bigger context window?
Claude Haiku 4.5 does, with 200K tokens against 164K.
How many benchmarks do Claude Haiku 4.5 and DeepSeek-V3.2-Exp share?
38 benchmarks have published results for both models. Claude Haiku 4.5 has 53 scored results on Noometry and DeepSeek-V3.2-Exp has 48.