Model comparison
Claude 3.5 Haiku vs DeepSeek V4 Pro
DeepSeek V4 Pro is the stronger model overall, scoring 54.2 to 28.9 on the Noometry Index.
Last verified . 25 shared benchmarks.
DeepSeekSummary
- They share 25 benchmarks with published results for both. Claude 3.5 Haiku scores higher in 0 categories and DeepSeek V4 Pro in 9 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek V4 Pro leads 64.6 to 13.7.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 4.3% for Claude 3.5 Haiku and 98.6% for DeepSeek V4 Pro.
- DeepSeek V4 Pro has downloadable open weights; the other is API-only.
Side by side
| Claude 3.5 Haiku | DeepSeek V4 Pro | |
|---|---|---|
| Provider | Anthropic | DeepSeek |
| Noometry Index | 28.9 | 54.2 |
| Released | 2024-10-22 | 2026-04-24 |
| Weights | Proprietary | Open |
| Context window | — | 1M |
| Max output | — | 393K |
| Input $ / M tokens | — | $0.43 |
| Output $ / M tokens | — | $0.87 |
| Results tracked | 49 | 48 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek V4 Pro leads
Claude 3.5 Haiku: 32.6 (#258), DeepSeek V4 Pro: 52.2 (#31)
| Benchmark | Claude 3.5 Haiku | DeepSeek V4 Pro |
|---|---|---|
| SciCode | 27.4% | 51% |
| WeirdML | 30.7% | 66.2% |
| LMArena Coding | 1287 | 1470 |
| SWE-bench Verified | — | 77.6% |
| FrontierCode | — | 28.6% |
| Aider Polyglot | 28% | — |
| LMArena WebDev | — | 1582 |
| BigCodeBench Instruct | 46.1% | — |
| LiveBench Coding | 51.4% | — |
| BigCodeBench Complete | 59% | — |
| CadEval | 32% | — |
| ALE-Bench | — | 1,403 |
Agentic & Tool Use DeepSeek V4 Pro leads
Claude 3.5 Haiku: 27.7 (#90), DeepSeek V4 Pro: 32.7 (#54)
| Benchmark | Claude 3.5 Haiku | DeepSeek V4 Pro |
|---|---|---|
| APEX-Agents | — | 47.3% |
| BALROG | 19.3% | — |
| Vending-Bench 2 | — | 3,285 |
Reasoning DeepSeek V4 Pro leads
Claude 3.5 Haiku: 17.1 (#286), DeepSeek V4 Pro: 56.4 (#22)
| Benchmark | Claude 3.5 Haiku | DeepSeek V4 Pro |
|---|---|---|
| CritPt | 0% | 18% |
| LMArena Hard Prompts | 1251 | 1461 |
| DTBench | 56.7% | 93.9% |
| Epoch Capabilities Index | 127.15 | 155.39 |
| ARC-AGI-2 | — | 61.3% |
| Kagi LLM Benchmark | — | 53.5% |
| NYT Connections (extended) | — | 91.3% |
| ARC-AGI-1 | — | 90.5% |
| Chess Puzzles | — | 47% |
| LiveBench Reasoning | 28.1% | — |
| Mystery Game Puzzles | — | 43% |
| LiveBench Data Analysis | 48.5% | — |
| LMCA | — | 45.5% |
| Surface Evolver Bench | — | 40% |
| ForecastBench | — | 56.1 |
| LiveBench | 43.5% | — |
Math DeepSeek V4 Pro leads
Claude 3.5 Haiku: 13.7 (#294), DeepSeek V4 Pro: 64.6 (#24)
| Benchmark | Claude 3.5 Haiku | DeepSeek V4 Pro |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 4.3% | 98.6% |
| LMArena Math | 1244 | 1455 |
| FrontierMath (Tiers 1-3) | — | 64.6% |
| FrontierMath Tier 4 | — | 26.8% |
| MathArena Final-Answer Competitions | — | 76.6% |
| ProofBench | — | 50% |
| Omni-MATH | 22.4% | — |
| LiveBench Math | 35.5% | — |
| MATH Level 5 | 46.4% | — |
| FrontierMath (Feb 2025 set) | 0.3% | — |
Knowledge DeepSeek V4 Pro leads
Claude 3.5 Haiku: 18.7 (#276), DeepSeek V4 Pro: 59.5 (#28)
| Benchmark | Claude 3.5 Haiku | DeepSeek V4 Pro |
|---|---|---|
| GPQA Diamond | 38.1% | 91.7% |
| LMArena Expert | 1208 | 1463 |
| SimpleQA Verified | — | 52.9% |
| MMLU-Pro | 60.5% | — |
| Confabulations | 36.7% | — |
| Vectara Hallucination Rate | — | 8.6% |
| GPQA (HELM) | 36.3% | — |
| MMLU | 74.3% | — |
Multimodal Not comparable
Claude 3.5 Haiku: 26.3 (#113), DeepSeek V4 Pro: —
| Benchmark | Claude 3.5 Haiku | DeepSeek V4 Pro |
|---|---|---|
| LMArena Vision | 1092 | — |
| GeoBench | 34% | — |
Multilingual DeepSeek V4 Pro leads
Claude 3.5 Haiku: 40.2 (#213), DeepSeek V4 Pro: 54.5 (#44)
| Benchmark | Claude 3.5 Haiku | DeepSeek V4 Pro |
|---|---|---|
| LMArena Non-English | 1238 | 1438 |
| LMArena Chinese | 1229 | 1486 |
| LMArena French | 1262 | 1473 |
| LMArena German | 1237 | 1458 |
| LMArena Japanese | 1175 | 1448 |
| LMArena Korean | 1172 | 1452 |
| LMArena Russian | 1253 | 1452 |
| LMArena Spanish | 1261 | 1457 |
Instruction Following DeepSeek V4 Pro leads
Claude 3.5 Haiku: 63.1 (#229), DeepSeek V4 Pro: 76.3 (#44)
| Benchmark | Claude 3.5 Haiku | DeepSeek V4 Pro |
|---|---|---|
| LMArena Instruction Following | 1241 | 1448 |
| LiveBench Instruction Following | 61.9% | — |
| IFEval | 79.2% | — |
Long Context DeepSeek V4 Pro leads
Claude 3.5 Haiku: 38.3 (#195), DeepSeek V4 Pro: 45.0 (#49)
| Benchmark | Claude 3.5 Haiku | DeepSeek V4 Pro |
|---|---|---|
| LMArena Longer Query | 1261 | 1459 |
| CL-bench Life | — | 13.5% |
Writing & Preference DeepSeek V4 Pro leads
Claude 3.5 Haiku: 42.9 (#228), DeepSeek V4 Pro: 65.6 (#45)
| Benchmark | Claude 3.5 Haiku | DeepSeek V4 Pro |
|---|---|---|
| LMArena Text | 1255 | 1451 |
| LMArena Creative Writing | 1233 | 1446 |
| EQ-Bench Creative Writing | 1146 | 1553 |
| LMArena Multi-Turn | 1265 | 1467 |
| Short-Story Creative Writing | 73.5% | — |
| WildBench | 76% | — |
| EQ-Bench 4 | — | 1166 |
| LiveBench Language | 35.4% | — |
Frequently asked questions
Is Claude 3.5 Haiku better than DeepSeek V4 Pro?
DeepSeek V4 Pro is the stronger model overall, scoring 54.2 to 28.9 on the Noometry Index.
Is Claude 3.5 Haiku or DeepSeek V4 Pro better for coding?
DeepSeek V4 Pro scores higher on coding benchmarks: 52.2 versus 32.6 in the Noometry coding category.
How many benchmarks do Claude 3.5 Haiku and DeepSeek V4 Pro share?
25 benchmarks have published results for both models. Claude 3.5 Haiku has 49 scored results on Noometry and DeepSeek V4 Pro has 48.