GLM 5.2 vs Phi 4
GLM 5.2 leads Phi 4 on the Artificial Analysis Intelligence Index, 33.7 to 5.9, and Phi 4 costs 31.4x less per output token ($0.14 vs $4.40 per million). Updated October 22, 2026; 37 benchmarks have a result for both.
GLM 5.2 or Phi 4: which to pick
Only GLM 5.2 has a coding score, 61.1 on the Vibe Coding Index; Phi 4 has none, so the two cannot be ranked on coding.
Pick Phi 4 for output tokens 31.4x cheaper ($0.14 vs $4.40 per million).
- Shared results
- 37
- GLM 5.2 only
- 61
- Phi 4 only
- 9
- Like-for-like categories
- 7 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
No call
Only GLM 5.2 has a coding score, so no call is made.
One side not measured
Intelligence
AA Intelligence Index
GLM 5.2
GLM 5.2 leads, 33.7 against 5.9.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only GLM 5.2 has a AA Agentic Index result, 38.4.
Directional only
Reasoning
GPQA Diamond
No call
Only GLM 5.2 has a GPQA Diamond result, 85.6%.
Directional only
Knowledge
MMLU Pro
No call
Only GLM 5.2 has a MMLU Pro result, 86.7%.
Directional only
Web design
Design Arena: websites
No call
Only GLM 5.2 has a Design Arena: websites result, 1285.
Directional only
Long documents
Listed context window
GLM 5.2
GLM 5.2 takes 1.05M tokens against 16K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Phi 4
GLM 5.2 $0.0036, Phi 4 $0.0001 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only GLM 5.2 has the number, $0.86.
Directional only
At a glance
- Coding index
- GLM 5.2 61.1 from 4 benchmarks, Phi 4 not benchmarked.
- Widest category
- Agentic work, 99.1 points apart on average over 1 shared benchmark.
- Widest single gap
- τ²-Bench Telecom: GLM 5.2 99.1% against Phi 4 0.0%, 99.1 points apart on a 0 to 100 scale.
- Output price
- GLM 5.2 $4.40 against Phi 4 $0.14 per million output tokens, 31.4x apart.
- Context window
- GLM 5.2 takes 1.05M tokens, Phi 4 16K.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | GLM 5.2 | Bars | Phi 4 | Ties |
|---|---|---|---|---|
| Coding | 3 | 0 | ||
| Intelligence | 2 | 0 | ||
| Agentic work | 1 | 0 | ||
| Reasoning | 6 | 0 | ||
| Math | 3 | 0 | ||
| Knowledge | 4 | 0 | ||
| General | 18 | 0 | ||
| All shared boards | 37 | 0 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index30
| Benchmark | GLM 5.2 | Phi 4 | Leader |
|---|---|---|---|
| LMArena WebDev | 1604 | Not measured | Only GLM 5.2 measured |
| AA Coding Index | 68.8 | Not measured | Only GLM 5.2 measured |
| LiveCodeBench (Vals AI) | 69.5% | Not measured | Only GLM 5.2 measured |
| SWE-bench Verified (Vals AI) | 82.8% | Not measured | Only GLM 5.2 measured |
| Vibe Code Bench (Vals AI) | 64.0% | Not measured | Only GLM 5.2 measured |
| Code migration (Vals AI) | 37.9% | Not measured | Only GLM 5.2 measured |
| ProgramBench (Vals AI) | 0.5% | Not measured | Only GLM 5.2 measured |
| Terminal-Bench 2.1 (Vals AI) | 67.8% | Not measured | Only GLM 5.2 measured |
| SkillsBench (Vals AI) | 45.1% | Not measured | Only GLM 5.2 measured |
| LiveCodeBench (Artificial Analysis run) | Not measured | 23.1% | Only Phi 4 measured |
| SciCode | 51.2% | Not measured | Only GLM 5.2 measured |
| Terminal-Bench Hard (Artificial Analysis run) | 50.8% | 3.8% | GLM 5.2 |
| FrontierCode | 24.5% | Not measured | Only GLM 5.2 measured |
| WeirdML | 70.1% | Not measured | Only GLM 5.2 measured |
| ALE-Bench | 1047.0 | Not measured | Only GLM 5.2 measured |
| LMArena Text: coding | 1487 | 1231 | GLM 5.2 |
| LMArena Text: software and IT | 1486 | 1228 | GLM 5.2 |
| LMArena WebDev: React | 1617 | Not measured | Only GLM 5.2 measured |
| LMArena WebDev: HTML | 1565 | Not measured | Only GLM 5.2 measured |
| Neither model has a result on the other 20 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | GLM 5.2 | Phi 4 | Leader |
|---|---|---|---|
| AA Intelligence Index | 33.7 | 5.9 | GLM 5.2 |
| Epoch Capabilities Index | 151.9 | 130.5 | GLM 5.2 |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | GLM 5.2 | Phi 4 | Leader |
|---|---|---|---|
| AA Agentic Index | 38.4 | Not measured | Only GLM 5.2 measured |
| Design Arena: agents building web apps | 1222 | Not measured | Only GLM 5.2 measured |
| Design Arena: agents building full-stack apps | 1241 | Not measured | Only GLM 5.2 measured |
| Design Arena: agents building mobile apps | 1159 | Not measured | Only GLM 5.2 measured |
| Design Arena: agents building Android apps | 1178 | Not measured | Only GLM 5.2 measured |
| Design Arena: agents building Godot games | 1142 | Not measured | Only GLM 5.2 measured |
| Design Arena: agents building games | 1206 | Not measured | Only GLM 5.2 measured |
| Design Arena: HTML slides | 1125 | Not measured | Only GLM 5.2 measured |
| Design Arena: PowerPoint slides | 1170 | Not measured | Only GLM 5.2 measured |
| SREBench (Vals AI) | 0.0% | Not measured | Only GLM 5.2 measured |
| Finance Agent (Vals AI) | 49.7% | Not measured | Only GLM 5.2 measured |
| Legal Research Bench (Vals AI) | 31.3% | Not measured | Only GLM 5.2 measured |
| Excel Modeling Benchmark (Vals AI) | 61.5% | Not measured | Only GLM 5.2 measured |
| τ²-Bench Telecom | 99.1% | 0.0% | GLM 5.2 |
| APEX-Agents | 45.2% | Not measured | Only GLM 5.2 measured |
| BALROG | Not measured | 11.6% | Only Phi 4 measured |
| PostTrainBench | 31.7% | Not measured | Only GLM 5.2 measured |
| Vending-Bench 2 | $8,314 | Not measured | Only GLM 5.2 measured |
| Berkeley Function Calling Leaderboard (BFCL) | Not measured | 28.8% | Only Phi 4 measured |
| Neither model has a result on the other 22 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index60
| Benchmark | GLM 5.2 | Phi 4 | Leader |
|---|---|---|---|
| GPQA Diamond | 85.6% | Not measured | Only GLM 5.2 measured |
| GPQA Diamond (Artificial Analysis run) | 89.5% | 57.5% | GLM 5.2 |
| Humanity's Last Exam (Artificial Analysis run) | 41.1% | 3.8% | GLM 5.2 |
| GPQA Diamond (Epoch AI run) | 91.9% | 56.1% | GLM 5.2 |
| ARC-AGI | 77.0% | Not measured | Only GLM 5.2 measured |
| ARC-AGI-2 | 22.8% | Not measured | Only GLM 5.2 measured |
| SimpleBench | 58.8% | Not measured | Only GLM 5.2 measured |
| Chess puzzles | 22.0% | Not measured | Only GLM 5.2 measured |
| Mystery game puzzles | 19.0% | Not measured | Only GLM 5.2 measured |
| EBR-bench | 9.5% | Not measured | Only GLM 5.2 measured |
| LMArena Text: hard prompts | 1481 | 1220 | GLM 5.2 |
| LMArena Text: expert | 1487 | 1203 | GLM 5.2 |
| LMArena Text: hard prompts in English | 1481 | 1230 | GLM 5.2 |
| Kagi LLM Benchmark | 62.6% | Not measured | Only GLM 5.2 measured |
| Neither model has a result on the other 8 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index30
| Benchmark | GLM 5.2 | Phi 4 | Leader |
|---|---|---|---|
| AA Math Index | Not measured | 18.0 | Only Phi 4 measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 18.0% | Only Phi 4 measured |
| OTIS Mock AIME | 86.4% | 13.8% | GLM 5.2 |
| FrontierMath, tiers 1 to 3 | 59.2% | Not measured | Only GLM 5.2 measured |
| FrontierMath, tier 4 | 29.3% | Not measured | Only GLM 5.2 measured |
| ProofBench | 35.0% | Not measured | Only GLM 5.2 measured |
| MATH level 5 | Not measured | 64.9% | Only Phi 4 measured |
| LMArena Text: math | 1483 | 1246 | GLM 5.2 |
| LMArena Text: mathematical work | 1489 | 1254 | GLM 5.2 |
| Neither model has a result on the other 4 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index40
| Benchmark | GLM 5.2 | Phi 4 | Leader |
|---|---|---|---|
| MMLU Pro | 86.7% | Not measured | Only GLM 5.2 measured |
| LegalBench (Vals AI) | 84.1% | Not measured | Only GLM 5.2 measured |
| CorpFin (Vals AI) | 66.1% | Not measured | Only GLM 5.2 measured |
| TaxEval (Vals AI) | 73.3% | Not measured | Only GLM 5.2 measured |
| MedCode (Vals AI) | 40.8% | Not measured | Only GLM 5.2 measured |
| MMLU Pro (Artificial Analysis run) | Not measured | 71.4% | Only Phi 4 measured |
| SimpleQA Verified | 34.2% | Not measured | Only GLM 5.2 measured |
| MMLU | Not measured | 84.8% | Only Phi 4 measured |
| LMArena Text: science | 1489 | 1206 | GLM 5.2 |
| LMArena Text: medicine and healthcare | 1475 | 1202 | GLM 5.2 |
| LMArena Text: legal and government | 1486 | 1241 | GLM 5.2 |
| LMArena Text: business and finance | 1451 | 1199 | GLM 5.2 |
| Neither model has a result on the other 6 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | GLM 5.2 | Phi 4 | Leader |
|---|---|---|---|
| Design Arena: websites | 1285 | Not measured | Only GLM 5.2 measured |
| Design Arena: UI components | 1294 | Not measured | Only GLM 5.2 measured |
| Design Arena: game dev | 1268 | Not measured | Only GLM 5.2 measured |
| Design Arena: data viz | 1285 | Not measured | Only GLM 5.2 measured |
| Design Arena: 3D | 1287 | Not measured | Only GLM 5.2 measured |
| Design Arena: all code categories | 1287 | Not measured | Only GLM 5.2 measured |
| Design Arena: SVG | 1213 | Not measured | Only GLM 5.2 measured |
| Design Arena: ASCII art | 1191 | Not measured | Only GLM 5.2 measured |
| Neither model has a result on the other 1 board here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index180
| Benchmark | GLM 5.2 | Phi 4 | Leader |
|---|---|---|---|
| MedScribe (Vals AI) | 83.5% | Not measured | Only GLM 5.2 measured |
| IFBench | 73.3% | 23.5% | GLM 5.2 |
| Long context reasoning (AA-LCR) | 78.3% | 0.0% | GLM 5.2 |
| DTBench | 93.6% | Not measured | Only GLM 5.2 measured |
| LMCA | 45.8% | Not measured | Only GLM 5.2 measured |
| Creative writing (Lech Mazur) | Not measured | 62.6% | Only Phi 4 measured |
| Surface Evolver Bench | 55.6% | Not measured | Only GLM 5.2 measured |
| LMArena Text | 1470 | 1217 | GLM 5.2 |
| LMArena Text: instruction following | 1466 | 1201 | GLM 5.2 |
| LMArena Text: creative writing | 1463 | 1181 | GLM 5.2 |
| LMArena Text: English | 1475 | 1231 | GLM 5.2 |
| LMArena Text: non-English | 1459 | 1197 | GLM 5.2 |
| LMArena Text: Chinese | 1522 | 1212 | GLM 5.2 |
| LMArena Text: German | 1470 | 1222 | GLM 5.2 |
| LMArena Text: Japanese | 1450 | 1158 | GLM 5.2 |
| LMArena Text: Korean | 1440 | 1151 | GLM 5.2 |
| LMArena Text: French | 1479 | 1224 | GLM 5.2 |
| LMArena Text: Spanish | 1478 | 1234 | GLM 5.2 |
| LMArena Text: Russian | 1468 | 1209 | GLM 5.2 |
| LMArena Text: Polish | 1456 | Not measured | Only GLM 5.2 measured |
| LMArena Text: multi-turn | 1469 | 1206 | GLM 5.2 |
| LMArena Text: longer queries | 1480 | 1217 | GLM 5.2 |
| LMArena Text: writing and language | 1463 | 1198 | GLM 5.2 |
| LMArena Text: entertainment and media | 1450 | 1162 | GLM 5.2 |
| Neither model has a result on the other 22 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GLM 5.2 | Phi 4 | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Phi 4 costs less
Repository review
50K input + 3K output tokens
Phi 4 costs less
Agent loop
200K input + 10K output tokens
Phi 4 costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | Z.ai GLM 5.2 | Microsoft Phi 4 |
|---|---|---|
| Maker | Z.ai | Microsoft |
| Weights | Open weights | Open weights |
| Released | June 16, 2026 | January 10, 2025 |
| Context window | 1.05M | 16K |
| Input $/Mtok | $1.40 | $0.07 |
| Output $/Mtok | $4.40 | $0.14 |
| Coding index | 61.1 | Not benchmarked |
| Independent measurements behind it | 4 | 0 |
| Cost per solved task | $0.86 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GLM 5.2 better than Phi 4 for coding?
Only GLM 5.2 has a coding score, 61.1 on the Vibe Coding Index; Phi 4 has none, so the two cannot be ranked on coding. Pick Phi 4 for output tokens 31.4x cheaper ($0.14 vs $4.40 per million).
Which is cheaper, GLM 5.2 or Phi 4?
GLM 5.2 costs $1.40 per million input tokens and $4.40 per million output tokens; Phi 4 costs $0.07 and $0.14. Cost per solved task is known for GLM 5.2 only.
Which has the bigger context window, GLM 5.2 or Phi 4?
GLM 5.2, with 1.05M tokens against 16K.
How many benchmarks is the GLM 5.2 vs Phi 4 comparison based on?
37 benchmarks have a result for both models as of October 22, 2026. GLM 5.2's coding index rests on 4 independent benchmarks and Phi 4's on 0. Where only one model has a result, the page says so instead of naming a winner.