Model comparison
DeepSeek-V2.5 (Sep 2024) vs DeepSeek V4 Pro
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 37.6 on the Noometry Index.
Last verified . 17 shared benchmarks.
DeepSeekSummary
- They share 17 benchmarks with published results for both. DeepSeek-V2.5 (Sep 2024) scores higher in 0 categories and DeepSeek V4 Pro in 8 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek V4 Pro leads 56.5 to 25.6.
Side by side
| DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Pro | |
|---|---|---|
| Provider | DeepSeek | DeepSeek |
| Noometry Index | 37.6 | 54.3 |
| Released | 2024-09-06 | 2026-04-24 |
| Weights | Open | Open |
| Context window | — | 1M |
| Max output | — | 393K |
| Input $ / M tokens | — | $0.66 |
| Output $ / M tokens | — | $1.98 |
| Results tracked | 22 | 48 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek V4 Pro leads
DeepSeek-V2.5 (Sep 2024): 31.7 (#281), DeepSeek V4 Pro: 52.4 (#34)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Pro |
|---|---|---|
| LMArena Coding | 1309 | 1470 |
| SWE-bench Verified | — | 77.6% |
| FrontierCode | — | 28.6% |
| Aider Polyglot | 17.8% | — |
| LMArena WebDev | — | 1582 |
| SciCode | — | 51% |
| WeirdML | — | 66.2% |
| BigCodeBench Instruct | 48.6% | — |
| BigCodeBench Complete | 53.2% | — |
| ALE-Bench | — | 1,403 |
| HumanEval+ | 83.5% | — |
| MBPP+ | 74.1% | — |
Agentic & Tool Use Not comparable
DeepSeek-V2.5 (Sep 2024): —, DeepSeek V4 Pro: 32.8 (#58)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Pro |
|---|---|---|
| APEX-Agents | — | 47.3% |
| Vending-Bench 2 | — | 3,285 |
Reasoning DeepSeek V4 Pro leads
DeepSeek-V2.5 (Sep 2024): 25.6 (#145), DeepSeek V4 Pro: 56.5 (#24)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Pro |
|---|---|---|
| LMArena Hard Prompts | 1289 | 1461 |
| ARC-AGI-2 | — | 61.3% |
| Kagi LLM Benchmark | — | 53.5% |
| NYT Connections (extended) | — | 91.3% |
| ARC-AGI-1 | — | 90.5% |
| CritPt | — | 18% |
| Chess Puzzles | — | 47% |
| Mystery Game Puzzles | — | 43% |
| DTBench | — | 93.9% |
| LMCA | — | 45.5% |
| Surface Evolver Bench | — | 40% |
| Epoch Capabilities Index | — | 155.31 |
| ForecastBench | — | 56.1 |
Math DeepSeek V4 Pro leads
DeepSeek-V2.5 (Sep 2024): 35.9 (#177), DeepSeek V4 Pro: 64.8 (#30)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Pro |
|---|---|---|
| LMArena Math | 1288 | 1455 |
| FrontierMath (Tiers 1-3) | — | 64.6% |
| FrontierMath Tier 4 | — | 26.8% |
| MathArena Final-Answer Competitions | — | 76.6% |
| OTIS Mock AIME 2024-2025 | — | 98.6% |
| ProofBench | — | 50% |
Knowledge DeepSeek V4 Pro leads
DeepSeek-V2.5 (Sep 2024): 34.8 (#193), DeepSeek V4 Pro: 59.5 (#31)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Pro |
|---|---|---|
| LMArena Expert | 1266 | 1464 |
| GPQA Diamond | — | 91.7% |
| SimpleQA Verified | — | 52.9% |
| Vectara Hallucination Rate | — | 8.6% |
Multilingual DeepSeek V4 Pro leads
DeepSeek-V2.5 (Sep 2024): 42.5 (#193), DeepSeek V4 Pro: 54.4 (#45)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Pro |
|---|---|---|
| LMArena Non-English | 1273 | 1439 |
| LMArena Chinese | 1318 | 1486 |
| LMArena French | 1289 | 1472 |
| LMArena German | 1258 | 1458 |
| LMArena Japanese | 1228 | 1445 |
| LMArena Korean | 1209 | 1447 |
| LMArena Russian | 1289 | 1453 |
| LMArena Spanish | 1248 | 1458 |
Instruction Following DeepSeek V4 Pro leads
DeepSeek-V2.5 (Sep 2024): 67.5 (#194), DeepSeek V4 Pro: 76.1 (#47)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Pro |
|---|---|---|
| LMArena Instruction Following | 1280 | 1448 |
Long Context DeepSeek V4 Pro leads
DeepSeek-V2.5 (Sep 2024): 39.5 (#174), DeepSeek V4 Pro: 45.0 (#51)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Pro |
|---|---|---|
| LMArena Longer Query | 1301 | 1458 |
| CL-bench Life | — | 13.5% |
Writing & Preference DeepSeek V4 Pro leads
DeepSeek-V2.5 (Sep 2024): 49.8 (#187), DeepSeek V4 Pro: 65.5 (#46)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Pro |
|---|---|---|
| LMArena Text | 1294 | 1451 |
| LMArena Creative Writing | 1285 | 1446 |
| LMArena Multi-Turn | 1297 | 1467 |
| EQ-Bench Creative Writing | — | 1553 |
| EQ-Bench 4 | — | 1166 |
Frequently asked questions
Is DeepSeek-V2.5 (Sep 2024) better than DeepSeek V4 Pro?
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 37.6 on the Noometry Index.
Is DeepSeek-V2.5 (Sep 2024) or DeepSeek V4 Pro better for coding?
DeepSeek V4 Pro scores higher on coding benchmarks: 52.4 versus 31.7 in the Noometry coding category.
How many benchmarks do DeepSeek-V2.5 (Sep 2024) and DeepSeek V4 Pro share?
17 benchmarks have published results for both models. DeepSeek-V2.5 (Sep 2024) has 22 scored results on Noometry and DeepSeek V4 Pro has 48.