Model comparison
GLM-5.2 vs Phi-4
GLM-5.2 is the stronger model overall, scoring 51.1 to 31.2 on the Noometry Index. Phi-4 costs 25× less per token, which makes it the better buy when GLM-5.2's lead doesn't matter for your workload.
Last verified . 21 shared benchmarks.
MicrosoftSummary
- They share 21 benchmarks with published results for both. GLM-5.2 scores higher in 9 categories and Phi-4 in 0 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where GLM-5.2 leads 55.7 to 20.8.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 86.4% for GLM-5.2 and 13.8% for Phi-4.
- Phi-4 is cheaper at $0.07 / $0.14 per million input/output tokens, against $1.40 / $4.40 for GLM-5.2.
- GLM-5.2 accepts more context: 1M tokens versus 128K.
Side by side
| GLM-5.2 | Phi-4 | |
|---|---|---|
| Provider | Z.ai (Zhipu) | Microsoft |
| Noometry Index | 51.1 | 31.2 |
| Released | 2026-06-13 | 2024-12-11 |
| Weights | Open | Open |
| Context window | 1M | 128K |
| Max output | 131K | 4K |
| Input $ / M tokens | $1.40 | $0.07 |
| Output $ / M tokens | $4.40 | $0.14 |
| Results tracked | 51 | 37 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GLM-5.2 leads
GLM-5.2: 51.3 (#41), Phi-4: 34.4 (#239)
| Benchmark | GLM-5.2 | Phi-4 |
|---|---|---|
| LMArena Coding | 1485 | 1231 |
| SWE-bench Verified | 78.7% | — |
| DeepSWE | 43.8% | — |
| FrontierCode | 24.5% | — |
| LMArena WebDev | 1603 | — |
| SciCode | 50.5% | — |
| WeirdML | 70.1% | — |
| BigCodeBench Instruct | — | 45.5% |
| LiveBench Coding | — | 30.7% |
| BigCodeBench Complete | — | 55.4% |
| ALE-Bench | 1,047 | — |
Agentic & Tool Use GLM-5.2 leads
GLM-5.2: 32.4 (#63), Phi-4: 22.8 (#128)
| Benchmark | GLM-5.2 | Phi-4 |
|---|---|---|
| APEX-Agents | 45.2% | — |
| Berkeley Function Calling Leaderboard | — | 28.8% |
| τ²-bench Banking | 37.1% | — |
| PostTrainBench | 31.7% | — |
| BALROG | — | 11.6% |
| GBAEval | 0% | — |
| Vending-Bench 2 | 8,314 | — |
Reasoning GLM-5.2 leads
GLM-5.2: 42.3 (#52), Phi-4: 17.7 (#291)
| Benchmark | GLM-5.2 | Phi-4 |
|---|---|---|
| Chess Puzzles | 21% | 1% |
| LMArena Hard Prompts | 1480 | 1220 |
| Epoch Capabilities Index | 151.78 | 130.42 |
| ARC-AGI-2 | 22.8% | — |
| SimpleBench | 58.8% | — |
| Kagi LLM Benchmark | 62.6% | — |
| NYT Connections (extended) | 74.3% | — |
| ARC-AGI-1 | 77% | — |
| CritPt | 20.9% | — |
| EBR-Bench | 9.5% | — |
| LiveBench Reasoning | — | 47.8% |
| Mystery Game Puzzles | 19% | — |
| DTBench | 93.6% | — |
| LiveBench Data Analysis | — | 45.2% |
| LMCA | 45.8% | — |
| Surface Evolver Bench | 55.6% | — |
| LiveBench | — | 41.6% |
Math GLM-5.2 leads
GLM-5.2: 55.7 (#43), Phi-4: 20.8 (#285)
| Benchmark | GLM-5.2 | Phi-4 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 86.4% | 13.8% |
| LMArena Math | 1482 | 1246 |
| FrontierMath (Tiers 1-3) | 59.2% | — |
| FrontierMath Tier 4 | 29.3% | — |
| MathArena Final-Answer Competitions | 67.6% | — |
| ProofBench | 35% | — |
| LiveBench Math | — | 42% |
| MATH Level 5 | — | 64.9% |
Knowledge GLM-5.2 leads
GLM-5.2: 57.1 (#40), Phi-4: 32.6 (#209)
| Benchmark | GLM-5.2 | Phi-4 |
|---|---|---|
| GPQA Diamond | 91.9% | 56.1% |
| LMArena Expert | 1486 | 1203 |
| SimpleQA Verified | 34.2% | — |
| Confabulations | — | 29.4% |
| Vectara Hallucination Rate | — | 3.7% |
| MMLU | — | 84.8% |
Multilingual GLM-5.2 leads
GLM-5.2: 55.8 (#26), Phi-4: 37.2 (#237)
| Benchmark | GLM-5.2 | Phi-4 |
|---|---|---|
| LMArena Non-English | 1459 | 1197 |
| LMArena Chinese | 1519 | 1212 |
| LMArena French | 1479 | 1224 |
| LMArena German | 1468 | 1222 |
| LMArena Japanese | 1451 | 1158 |
| LMArena Korean | 1445 | 1151 |
| LMArena Russian | 1466 | 1209 |
| LMArena Spanish | 1477 | 1234 |
Instruction Following GLM-5.2 leads
GLM-5.2: 76.9 (#34), Phi-4: 60.4 (#251)
| Benchmark | GLM-5.2 | Phi-4 |
|---|---|---|
| LMArena Instruction Following | 1465 | 1201 |
| LiveBench Instruction Following | — | 58.4% |
Long Context GLM-5.2 leads
GLM-5.2: 45.3 (#43), Phi-4: 36.9 (#226)
| Benchmark | GLM-5.2 | Phi-4 |
|---|---|---|
| LMArena Longer Query | 1479 | 1217 |
Writing & Preference GLM-5.2 leads
GLM-5.2: 70.4 (#21), Phi-4: 40.5 (#244)
| Benchmark | GLM-5.2 | Phi-4 |
|---|---|---|
| LMArena Text | 1470 | 1217 |
| LMArena Creative Writing | 1462 | 1182 |
| LMArena Multi-Turn | 1469 | 1206 |
| Short-Story Creative Writing | — | 62.6% |
| EQ-Bench Creative Writing | 1757 | — |
| EQ-Bench 4 | 1222 | — |
| LiveBench Language | — | 25.6% |
Frequently asked questions
Is GLM-5.2 better than Phi-4?
GLM-5.2 is the stronger model overall, scoring 51.1 to 31.2 on the Noometry Index. Phi-4 costs 25× less per token, which makes it the better buy when GLM-5.2's lead doesn't matter for your workload.
Which is cheaper, GLM-5.2 or Phi-4?
Phi-4 is cheaper. It lists at $0.07 per million input tokens and $0.14 per million output tokens; GLM-5.2 lists at $1.40 and $4.40.
Is GLM-5.2 or Phi-4 better for coding?
GLM-5.2 scores higher on coding benchmarks: 51.3 versus 34.4 in the Noometry coding category.
Which has the bigger context window?
GLM-5.2 does, with 1M tokens against 128K.
How many benchmarks do GLM-5.2 and Phi-4 share?
21 benchmarks have published results for both models. GLM-5.2 has 51 scored results on Noometry and Phi-4 has 37.