Model comparison
DeepSeek-V2.5 (Sep 2024) vs DeepSeek V4 Flash
DeepSeek V4 Flash is the stronger model overall, scoring 53.6 to 37.0 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 Flash in 8 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek V4 Flash leads 53.7 to 24.5.
Side by side
| DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Flash | |
|---|---|---|
| Provider | DeepSeek | DeepSeek |
| Noometry Index | 37.0 | 53.6 |
| Released | 2024-09-06 | 2026-04-24 |
| Weights | Open | Open |
| Context window | — | 1M |
| Max output | — | 393K |
| Input $ / M tokens | — | $0.15 |
| Output $ / M tokens | — | $0.60 |
| Results tracked | 22 | 41 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek V4 Flash leads
DeepSeek-V2.5 (Sep 2024): 31.3 (#274), DeepSeek V4 Flash: 47.6 (#56)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Flash |
|---|---|---|
| LMArena Coding | 1310 | 1457 |
| FrontierCode | — | 18.8% |
| Aider Polyglot | 17.8% | — |
| LMArena WebDev | — | 1581 |
| SciCode | — | 49.9% |
| WeirdML | — | 63% |
| BigCodeBench Instruct | 48.6% | — |
| BigCodeBench Complete | 53.2% | — |
| ALE-Bench | — | 1,306 |
| HumanEval+ | 83.5% | — |
| MBPP+ | 74.1% | — |
Reasoning DeepSeek V4 Flash leads
DeepSeek-V2.5 (Sep 2024): 24.5 (#145), DeepSeek V4 Flash: 53.7 (#25)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Flash |
|---|---|---|
| LMArena Hard Prompts | 1289 | 1443 |
| ARC-AGI-2 | — | 61.4% |
| SimpleBench | — | 61.1% |
| Kagi LLM Benchmark | — | 52.2% |
| NYT Connections (extended) | — | 89.6% |
| ARC-AGI-1 | — | 89% |
| CritPt | — | 16.6% |
| Chess Puzzles | — | 33% |
| Mystery Game Puzzles | — | 34% |
| DTBench | — | 90.9% |
| LMCA | — | 41.7% |
| Epoch Capabilities Index | — | 154.51 |
Math DeepSeek V4 Flash leads
DeepSeek-V2.5 (Sep 2024): 33.3 (#177), DeepSeek V4 Flash: 60.0 (#31)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Flash |
|---|---|---|
| LMArena Math | 1289 | 1427 |
| FrontierMath (Tiers 1-3) | — | 57.5% |
| FrontierMath Tier 4 | — | 24.4% |
| MathArena Final-Answer Competitions | — | 76.5% |
| OTIS Mock AIME 2024-2025 | — | 94.4% |
| ProofBench | — | 56% |
Knowledge DeepSeek V4 Flash leads
DeepSeek-V2.5 (Sep 2024): 34.2 (#190), DeepSeek V4 Flash: 55.5 (#43)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Flash |
|---|---|---|
| LMArena Expert | 1266 | 1441 |
| GPQA Diamond | — | 91% |
| SimpleQA Verified | — | 33.6% |
Multilingual DeepSeek V4 Flash leads
DeepSeek-V2.5 (Sep 2024): 42.7 (#188), DeepSeek V4 Flash: 53.2 (#71)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Flash |
|---|---|---|
| LMArena Non-English | 1273 | 1420 |
| LMArena Chinese | 1317 | 1467 |
| LMArena French | 1288 | 1438 |
| LMArena German | 1258 | 1418 |
| LMArena Japanese | 1228 | 1403 |
| LMArena Korean | 1209 | 1383 |
| LMArena Russian | 1289 | 1428 |
| LMArena Spanish | 1248 | 1437 |
Instruction Following DeepSeek V4 Flash leads
DeepSeek-V2.5 (Sep 2024): 67.7 (#189), DeepSeek V4 Flash: 75.1 (#76)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Flash |
|---|---|---|
| LMArena Instruction Following | 1280 | 1421 |
Long Context DeepSeek V4 Flash leads
DeepSeek-V2.5 (Sep 2024): 39.5 (#169), DeepSeek V4 Flash: 43.8 (#81)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Flash |
|---|---|---|
| LMArena Longer Query | 1301 | 1435 |
Writing & Preference DeepSeek V4 Flash leads
DeepSeek-V2.5 (Sep 2024): 50.0 (#182), DeepSeek V4 Flash: 64.0 (#59)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | DeepSeek V4 Flash |
|---|---|---|
| LMArena Text | 1294 | 1432 |
| LMArena Creative Writing | 1285 | 1403 |
| LMArena Multi-Turn | 1297 | 1449 |
| EQ-Bench Creative Writing | — | 1559 |
Frequently asked questions
Is DeepSeek-V2.5 (Sep 2024) better than DeepSeek V4 Flash?
DeepSeek V4 Flash is the stronger model overall, scoring 53.6 to 37.0 on the Noometry Index.
Is DeepSeek-V2.5 (Sep 2024) or DeepSeek V4 Flash better for coding?
DeepSeek V4 Flash scores higher on coding benchmarks: 47.6 versus 31.3 in the Noometry coding category.
How many benchmarks do DeepSeek-V2.5 (Sep 2024) and DeepSeek V4 Flash share?
17 benchmarks have published results for both models. DeepSeek-V2.5 (Sep 2024) has 22 scored results on Noometry and DeepSeek V4 Flash has 41.