Model comparison
Claude 3 Haiku vs DeepSeek V4 Pro
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 25.9 on the Noometry Index.
Last verified . 26 shared benchmarks.
DeepSeekSummary
- They share 26 benchmarks with published results for both. Claude 3 Haiku scores higher in 0 categories and DeepSeek V4 Pro in 8 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek V4 Pro leads 64.8 to 9.8.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 1.8% for Claude 3 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 Haiku | DeepSeek V4 Pro | |
|---|---|---|
| Provider | Anthropic | DeepSeek |
| Noometry Index | 25.9 | 54.3 |
| Released | 2024-03-07 | 2026-04-24 |
| Weights | Proprietary | Open |
| Context window | — | 1M |
| Max output | — | 393K |
| Input $ / M tokens | — | $0.66 |
| Output $ / M tokens | — | $1.98 |
| Results tracked | 37 | 48 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek V4 Pro leads
Claude 3 Haiku: 26.4 (#325), DeepSeek V4 Pro: 52.4 (#34)
| Benchmark | Claude 3 Haiku | DeepSeek V4 Pro |
|---|---|---|
| WeirdML | 9.8% | 66.2% |
| LMArena Coding | 1199 | 1470 |
| SWE-bench Verified | — | 77.6% |
| FrontierCode | — | 28.6% |
| LMArena WebDev | — | 1582 |
| SciCode | — | 51% |
| BigCodeBench Instruct | 39.4% | — |
| BigCodeBench Complete | 50.1% | — |
| CadEval | 12% | — |
| ALE-Bench | — | 1,403 |
| HumanEval+ | 68.9% | — |
| MBPP+ | 68.8% | — |
Agentic & Tool Use Not comparable
Claude 3 Haiku: —, DeepSeek V4 Pro: 32.8 (#58)
| Benchmark | Claude 3 Haiku | DeepSeek V4 Pro |
|---|---|---|
| APEX-Agents | — | 47.3% |
| Vending-Bench 2 | — | 3,285 |
Reasoning DeepSeek V4 Pro leads
Claude 3 Haiku: 16.3 (#307), DeepSeek V4 Pro: 56.5 (#24)
| Benchmark | Claude 3 Haiku | DeepSeek V4 Pro |
|---|---|---|
| Kagi LLM Benchmark | 34.2% | 53.5% |
| LMArena Hard Prompts | 1174 | 1461 |
| DTBench | 50.1% | 93.9% |
| LMCA | 8.8% | 45.5% |
| Epoch Capabilities Index | 118.35 | 155.31 |
| ForecastBench | 53.2 | 56.1 |
| ARC-AGI-2 | — | 61.3% |
| NYT Connections (extended) | — | 91.3% |
| ARC-AGI-1 | — | 90.5% |
| CritPt | — | 18% |
| Chess Puzzles | — | 47% |
| Mystery Game Puzzles | — | 43% |
| Surface Evolver Bench | — | 40% |
| WinoGrande | 74.2% | — |
Math DeepSeek V4 Pro leads
Claude 3 Haiku: 9.8 (#319), DeepSeek V4 Pro: 64.8 (#30)
| Benchmark | Claude 3 Haiku | DeepSeek V4 Pro |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 1.8% | 98.6% |
| LMArena Math | 1188 | 1455 |
| FrontierMath (Tiers 1-3) | — | 64.6% |
| FrontierMath Tier 4 | — | 26.8% |
| MathArena Final-Answer Competitions | — | 76.6% |
| ProofBench | — | 50% |
| MATH Level 5 | 14.9% | — |
Knowledge DeepSeek V4 Pro leads
Claude 3 Haiku: 17.3 (#285), DeepSeek V4 Pro: 59.5 (#31)
| Benchmark | Claude 3 Haiku | DeepSeek V4 Pro |
|---|---|---|
| GPQA Diamond | 36.3% | 91.7% |
| LMArena Expert | 1148 | 1464 |
| SimpleQA Verified | — | 52.9% |
| Confabulations | 34.2% | — |
| Vectara Hallucination Rate | — | 8.6% |
| MMLU | 73.8% | — |
Multimodal Not comparable
Claude 3 Haiku: 23.6 (#128), DeepSeek V4 Pro: —
| Benchmark | Claude 3 Haiku | DeepSeek V4 Pro |
|---|---|---|
| LMArena Vision | 950 | — |
| ScienceQA | 72% | — |
Multilingual DeepSeek V4 Pro leads
Claude 3 Haiku: 36.0 (#243), DeepSeek V4 Pro: 54.4 (#45)
| Benchmark | Claude 3 Haiku | DeepSeek V4 Pro |
|---|---|---|
| LMArena Non-English | 1178 | 1439 |
| LMArena Chinese | 1155 | 1486 |
| LMArena French | 1195 | 1472 |
| LMArena German | 1174 | 1458 |
| LMArena Japanese | 1102 | 1445 |
| LMArena Korean | 1109 | 1447 |
| LMArena Russian | 1204 | 1453 |
| LMArena Spanish | 1166 | 1458 |
Instruction Following DeepSeek V4 Pro leads
Claude 3 Haiku: 61.3 (#247), DeepSeek V4 Pro: 76.1 (#47)
| Benchmark | Claude 3 Haiku | DeepSeek V4 Pro |
|---|---|---|
| LMArena Instruction Following | 1173 | 1448 |
Long Context DeepSeek V4 Pro leads
Claude 3 Haiku: 36.1 (#237), DeepSeek V4 Pro: 45.0 (#51)
| Benchmark | Claude 3 Haiku | DeepSeek V4 Pro |
|---|---|---|
| LMArena Longer Query | 1190 | 1458 |
| CL-bench Life | — | 13.5% |
Writing & Preference DeepSeek V4 Pro leads
Claude 3 Haiku: 29.7 (#291), DeepSeek V4 Pro: 65.5 (#46)
| Benchmark | Claude 3 Haiku | DeepSeek V4 Pro |
|---|---|---|
| LMArena Text | 1195 | 1451 |
| LMArena Creative Writing | 1157 | 1446 |
| EQ-Bench Creative Writing | 717 | 1553 |
| LMArena Multi-Turn | 1190 | 1467 |
| EQ-Bench 4 | — | 1166 |
Frequently asked questions
Is Claude 3 Haiku better than DeepSeek V4 Pro?
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 25.9 on the Noometry Index.
Is Claude 3 Haiku or DeepSeek V4 Pro better for coding?
DeepSeek V4 Pro scores higher on coding benchmarks: 52.4 versus 26.4 in the Noometry coding category.
How many benchmarks do Claude 3 Haiku and DeepSeek V4 Pro share?
26 benchmarks have published results for both models. Claude 3 Haiku has 37 scored results on Noometry and DeepSeek V4 Pro has 48.