Model comparison
DeepSeek-V2.5 (Sep 2024) vs Phi-4
DeepSeek-V2.5 (Sep 2024) is the stronger model overall, scoring 37.6 to 31.2 on the Noometry Index.
Last verified . 19 shared benchmarks.
DeepSeek
MicrosoftSummary
- They share 19 benchmarks with published results for both. DeepSeek-V2.5 (Sep 2024) scores higher in 7 categories and Phi-4 in 1 category; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek-V2.5 (Sep 2024) leads 35.9 to 20.8.
Side by side
| DeepSeek-V2.5 (Sep 2024) | Phi-4 | |
|---|---|---|
| Provider | DeepSeek | Microsoft |
| Noometry Index | 37.6 | 31.2 |
| Released | 2024-09-06 | 2024-12-11 |
| Weights | Open | Open |
| Context window | — | 128K |
| Max output | — | 4K |
| Input $ / M tokens | — | $0.07 |
| Output $ / M tokens | — | $0.14 |
| Results tracked | 22 | 37 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Phi-4 leads
DeepSeek-V2.5 (Sep 2024): 31.7 (#281), Phi-4: 34.4 (#239)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Phi-4 |
|---|---|---|
| BigCodeBench Instruct | 48.6% | 45.5% |
| LMArena Coding | 1309 | 1231 |
| BigCodeBench Complete | 53.2% | 55.4% |
| Aider Polyglot | 17.8% | — |
| LiveBench Coding | — | 30.7% |
| HumanEval+ | 83.5% | — |
| MBPP+ | 74.1% | — |
Agentic & Tool Use Not comparable
DeepSeek-V2.5 (Sep 2024): —, Phi-4: 22.8 (#128)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Phi-4 |
|---|---|---|
| Berkeley Function Calling Leaderboard | — | 28.8% |
| BALROG | — | 11.6% |
Reasoning DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 25.6 (#145), Phi-4: 17.7 (#291)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Phi-4 |
|---|---|---|
| LMArena Hard Prompts | 1289 | 1220 |
| Chess Puzzles | — | 1% |
| LiveBench Reasoning | — | 47.8% |
| LiveBench Data Analysis | — | 45.2% |
| Epoch Capabilities Index | — | 130.42 |
| LiveBench | — | 41.6% |
Math DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 35.9 (#177), Phi-4: 20.8 (#285)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Phi-4 |
|---|---|---|
| LMArena Math | 1288 | 1246 |
| OTIS Mock AIME 2024-2025 | — | 13.8% |
| LiveBench Math | — | 42% |
| MATH Level 5 | — | 64.9% |
Knowledge DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 34.8 (#193), Phi-4: 32.6 (#209)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Phi-4 |
|---|---|---|
| LMArena Expert | 1266 | 1203 |
| GPQA Diamond | — | 56.1% |
| Confabulations | — | 29.4% |
| Vectara Hallucination Rate | — | 3.7% |
| MMLU | — | 84.8% |
Multilingual DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 42.5 (#193), Phi-4: 37.2 (#237)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Phi-4 |
|---|---|---|
| LMArena Non-English | 1273 | 1197 |
| LMArena Chinese | 1318 | 1212 |
| LMArena French | 1289 | 1224 |
| LMArena German | 1258 | 1222 |
| LMArena Japanese | 1228 | 1158 |
| LMArena Korean | 1209 | 1151 |
| LMArena Russian | 1289 | 1209 |
| LMArena Spanish | 1248 | 1234 |
Instruction Following DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 67.5 (#194), Phi-4: 60.4 (#251)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Phi-4 |
|---|---|---|
| LMArena Instruction Following | 1280 | 1201 |
| LiveBench Instruction Following | — | 58.4% |
Long Context DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 39.5 (#174), Phi-4: 36.9 (#226)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Phi-4 |
|---|---|---|
| LMArena Longer Query | 1301 | 1217 |
Writing & Preference DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 49.8 (#187), Phi-4: 40.5 (#244)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Phi-4 |
|---|---|---|
| LMArena Text | 1294 | 1217 |
| LMArena Creative Writing | 1285 | 1182 |
| LMArena Multi-Turn | 1297 | 1206 |
| Short-Story Creative Writing | — | 62.6% |
| LiveBench Language | — | 25.6% |
Frequently asked questions
Is DeepSeek-V2.5 (Sep 2024) better than Phi-4?
DeepSeek-V2.5 (Sep 2024) is the stronger model overall, scoring 37.6 to 31.2 on the Noometry Index.
Is DeepSeek-V2.5 (Sep 2024) or Phi-4 better for coding?
Phi-4 scores higher on coding benchmarks: 34.4 versus 31.7 in the Noometry coding category.
How many benchmarks do DeepSeek-V2.5 (Sep 2024) and Phi-4 share?
19 benchmarks have published results for both models. DeepSeek-V2.5 (Sep 2024) has 22 scored results on Noometry and Phi-4 has 37.