Model comparison
DeepSeek-V3.1-Terminus vs Phi-4
DeepSeek-V3.1-Terminus is the stronger model overall, scoring 43.1 to 31.2 on the Noometry Index. Phi-4 costs 5.2× less per token, which makes it the better buy when DeepSeek-V3.1-Terminus's lead doesn't matter for your workload.
Last verified . 10 shared benchmarks.
DeepSeek
MicrosoftSummary
- They share 10 benchmarks with published results for both. DeepSeek-V3.1-Terminus scores higher in 7 categories and Phi-4 in 0 categories; 7 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where DeepSeek-V3.1-Terminus leads 61.0 to 40.5.
- Phi-4 is cheaper at $0.07 / $0.14 per million input/output tokens, against $0.27 / $1 for DeepSeek-V3.1-Terminus.
- DeepSeek-V3.1-Terminus accepts more context: 164K tokens versus 128K.
Side by side
| DeepSeek-V3.1-Terminus | Phi-4 | |
|---|---|---|
| Provider | DeepSeek | Microsoft |
| Noometry Index | 43.1 | 31.2 |
| Released | 2025-09-22 | 2024-12-11 |
| Weights | Open | Open |
| Context window | 164K | 128K |
| Max output | 147K | 4K |
| Input $ / M tokens | $0.27 | $0.07 |
| Output $ / M tokens | $1 | $0.14 |
| Results tracked | 16 | 37 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3.1-Terminus leads
DeepSeek-V3.1-Terminus: 42.0 (#113), Phi-4: 34.4 (#239)
| Benchmark | DeepSeek-V3.1-Terminus | Phi-4 |
|---|---|---|
| LMArena Coding | 1426 | 1231 |
| SciCode | 40.6% | — |
| BigCodeBench Instruct | — | 45.5% |
| LiveBench Coding | — | 30.7% |
| BigCodeBench Complete | — | 55.4% |
| ALE-Bench | 745.17 | — |
Agentic & Tool Use Not comparable
DeepSeek-V3.1-Terminus: —, Phi-4: 22.8 (#128)
| Benchmark | DeepSeek-V3.1-Terminus | Phi-4 |
|---|---|---|
| Berkeley Function Calling Leaderboard | — | 28.8% |
| BALROG | — | 11.6% |
Reasoning DeepSeek-V3.1-Terminus leads
DeepSeek-V3.1-Terminus: 26.4 (#133), Phi-4: 17.7 (#291)
| Benchmark | DeepSeek-V3.1-Terminus | Phi-4 |
|---|---|---|
| LMArena Hard Prompts | 1426 | 1220 |
| Kagi LLM Benchmark | 57.4% | — |
| CritPt | 1.7% | — |
| Chess Puzzles | — | 1% |
| LiveBench Reasoning | — | 47.8% |
| DTBench | 81.3% | — |
| LiveBench Data Analysis | — | 45.2% |
| LMCA | 28.6% | — |
| Epoch Capabilities Index | — | 130.42 |
| LiveBench | — | 41.6% |
Math DeepSeek-V3.1-Terminus leads
DeepSeek-V3.1-Terminus: 38.5 (#137), Phi-4: 20.8 (#285)
| Benchmark | DeepSeek-V3.1-Terminus | Phi-4 |
|---|---|---|
| LMArena Math | 1402 | 1246 |
| OTIS Mock AIME 2024-2025 | — | 13.8% |
| LiveBench Math | — | 42% |
| MATH Level 5 | — | 64.9% |
Knowledge Not comparable
DeepSeek-V3.1-Terminus: —, Phi-4: 32.6 (#209)
| Benchmark | DeepSeek-V3.1-Terminus | Phi-4 |
|---|---|---|
| GPQA Diamond | — | 56.1% |
| Confabulations | — | 29.4% |
| Vectara Hallucination Rate | — | 3.7% |
| LMArena Expert | — | 1203 |
| MMLU | — | 84.8% |
Multilingual DeepSeek-V3.1-Terminus leads
DeepSeek-V3.1-Terminus: 52.1 (#92), Phi-4: 37.2 (#237)
| Benchmark | DeepSeek-V3.1-Terminus | Phi-4 |
|---|---|---|
| LMArena Non-English | 1407 | 1197 |
| LMArena Russian | 1436 | 1209 |
| LMArena Chinese | — | 1212 |
| LMArena French | — | 1224 |
| LMArena German | — | 1222 |
| LMArena Japanese | — | 1158 |
| LMArena Korean | — | 1151 |
| LMArena Spanish | — | 1234 |
Instruction Following DeepSeek-V3.1-Terminus leads
DeepSeek-V3.1-Terminus: 74.0 (#106), Phi-4: 60.4 (#251)
| Benchmark | DeepSeek-V3.1-Terminus | Phi-4 |
|---|---|---|
| LMArena Instruction Following | 1404 | 1201 |
| LiveBench Instruction Following | — | 58.4% |
Long Context DeepSeek-V3.1-Terminus leads
DeepSeek-V3.1-Terminus: 43.4 (#97), Phi-4: 36.9 (#226)
| Benchmark | DeepSeek-V3.1-Terminus | Phi-4 |
|---|---|---|
| LMArena Longer Query | 1421 | 1217 |
Writing & Preference DeepSeek-V3.1-Terminus leads
DeepSeek-V3.1-Terminus: 61.0 (#92), Phi-4: 40.5 (#244)
| Benchmark | DeepSeek-V3.1-Terminus | Phi-4 |
|---|---|---|
| LMArena Text | 1419 | 1217 |
| LMArena Creative Writing | 1403 | 1182 |
| LMArena Multi-Turn | 1411 | 1206 |
| Short-Story Creative Writing | — | 62.6% |
| LiveBench Language | — | 25.6% |
Frequently asked questions
Is DeepSeek-V3.1-Terminus better than Phi-4?
DeepSeek-V3.1-Terminus is the stronger model overall, scoring 43.1 to 31.2 on the Noometry Index. Phi-4 costs 5.2× less per token, which makes it the better buy when DeepSeek-V3.1-Terminus's lead doesn't matter for your workload.
Which is cheaper, DeepSeek-V3.1-Terminus or Phi-4?
Phi-4 is cheaper. It lists at $0.07 per million input tokens and $0.14 per million output tokens; DeepSeek-V3.1-Terminus lists at $0.27 and $1.
Is DeepSeek-V3.1-Terminus or Phi-4 better for coding?
DeepSeek-V3.1-Terminus scores higher on coding benchmarks: 42.0 versus 34.4 in the Noometry coding category.
Which has the bigger context window?
DeepSeek-V3.1-Terminus does, with 164K tokens against 128K.
How many benchmarks do DeepSeek-V3.1-Terminus and Phi-4 share?
10 benchmarks have published results for both models. DeepSeek-V3.1-Terminus has 16 scored results on Noometry and Phi-4 has 37.