GLM 5.3 vs GPT-5.4 Mini
GLM 5.3 leads GPT-5.4 Mini on coding, 70.6 to 35.8 on the Vibe Coding Index, at about the same output price ($4.40 vs $4.50 per million). Updated October 22, 2026; 66 benchmarks have a result for both.
GLM 5.3 or GPT-5.4 Mini: which to pick
Pick GLM 5.3 for coding: it scores 70.6 on the Vibe Coding Index against 35.8 for GPT-5.4 Mini.
Pick GPT-5.4 Mini for work measured by LiveCodeBench (Vals AI), where it leads.
- Shared results
- 66
- GLM 5.3 only
- 34
- GPT-5.4 Mini only
- 23
- 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
GLM 5.3
GLM 5.3 scores 70.6 against 35.8.
6 and 5 benchmarks behind the scores
Intelligence
AA Intelligence Index
GLM 5.3
GLM 5.3 leads, 44.8 against 24.1.
Same board, both measured
Agentic work
AA Agentic Index
GLM 5.3
GLM 5.3 leads, 53.1 against 17.9.
Same board, both measured
Reasoning
GPQA Diamond
GLM 5.3
GLM 5.3 leads, 88.1% against 83.1%.
Same board, both measured
Math
AIME
No call
Only GPT-5.4 Mini has a AIME result, 95.6%.
Directional only
Knowledge
MMLU Pro
GLM 5.3
GLM 5.3 leads, 86.8% against 84.5%.
Same board, both measured
Web design
Design Arena: websites
No call
Only GLM 5.3 has a Design Arena: websites result, 1295.
Directional only
General
Vals Index
No call
Only GLM 5.3 has a Vals Index result, 59.7%.
Directional only
Long documents
Listed context window
GLM 5.3
GLM 5.3 takes 1.05M tokens against 400K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
GPT-5.4 Mini
GLM 5.3 $0.0036, GPT-5.4 Mini $0.0030 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
GLM 5.3
GLM 5.3 $0.36, GPT-5.4 Mini $0.70.
Published attempt cost and pass rate
At a glance
- Coding index
- GLM 5.3 70.6 from 6 benchmarks, GPT-5.4 Mini 35.8 from 5 benchmarks.
- Widest category
- Agentic work, 24.8 points apart on average over 5 shared benchmarks.
- Widest single gap
- Terminal-Bench 4 (Vals AI): GLM 5.3 38.9% against GPT-5.4 Mini 2.5%, 48.5 points apart on a 0 to 100 scale.
- Output price
- GLM 5.3 $4.40 against GPT-5.4 Mini $4.50 per million output tokens, 1.0x apart.
- Context window
- GLM 5.3 takes 1.05M tokens, GPT-5.4 Mini 400K.
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 5.3 | Bars | GPT-5.4 Mini | Ties |
|---|---|---|---|---|
| Coding | 16 | 1 | ||
| Intelligence | 2 | 0 | ||
| Agentic work | 5 | 0 | ||
| Reasoning | 9 | 0 | ||
| Math | 6 | 0 | ||
| Knowledge | 7 | 0 | ||
| General | 20 | 0 | ||
| All shared boards | 65 | 1 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index161
| Benchmark | GLM 5.3 | GPT-5.4 Mini | Leader |
|---|---|---|---|
| LiveBench coding | 82.8% | Not measured | Only GLM 5.3 measured |
| LMArena WebDev | 1621 | 1397 | GLM 5.3 |
| AA Coding Index | 74.8 | 56.1 | GLM 5.3 |
| LiveCodeBench (Vals AI) | 80.5% | 81.5% | GPT-5.4 Mini |
| SWE-bench Verified (Vals AI) | 95.4% | 73.0% | GLM 5.3 |
| Terminal-Bench 4 (Vals AI) | 38.9% | 2.5% | GLM 5.3 |
| Vibe Code Bench (Vals AI) | 78.1% | 48.0% | GLM 5.3 |
| Code migration (Vals AI) | 44.2% | 12.9% | GLM 5.3 |
| ProgramBench (Vals AI) | 1.5% | 0.0% | GLM 5.3 |
| IOI (Vals AI) | 68.4% | Not measured | Only GLM 5.3 measured |
| Terminal-Bench 2.1 (Vals AI) | 71.5% | 54.7% | GLM 5.3 |
| Vibe Code Bench 1-100 (Vals AI) | 20.0% | Not measured | Only GLM 5.3 measured |
| SkillsBench (Vals AI) | 47.5% | Not measured | Only GLM 5.3 measured |
| CyberBench (Vals AI) | 72.1% | Not measured | Only GLM 5.3 measured |
| SciCode | 59.0% | 52.1% | GLM 5.3 |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 52.3% | Only GPT-5.4 Mini measured |
| FrontierCode | 40.1% | 27.0% | GLM 5.3 |
| FrontierSWE | 30.2% | Not measured | Only GLM 5.3 measured |
| WeirdML | 75.4% | 60.3% | GLM 5.3 |
| ALE-Bench | 1317.4 | 1188.6 | GLM 5.3 |
| LMArena Text: coding | 1497 | 1438 | GLM 5.3 |
| LMArena Text: software and IT | 1493 | 1436 | GLM 5.3 |
| LMArena WebDev: React | 1641 | 1386 | GLM 5.3 |
| LMArena WebDev: HTML | 1595 | 1419 | GLM 5.3 |
| Neither model has a result on the other 15 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index20
| Benchmark | GLM 5.3 | GPT-5.4 Mini | Leader |
|---|---|---|---|
| AA Intelligence Index | 44.8 | 24.1 | GLM 5.3 |
| Epoch Capabilities Index | 156.0 | 148.7 | GLM 5.3 |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index50
| Benchmark | GLM 5.3 | GPT-5.4 Mini | Leader |
|---|---|---|---|
| AA Agentic Index | 53.1 | 17.9 | GLM 5.3 |
| Design Arena: agents building mobile apps | 1194 | Not measured | Only GLM 5.3 measured |
| Design Arena: HTML slides | 1189 | Not measured | Only GLM 5.3 measured |
| Design Arena: PowerPoint slides | 1230 | Not measured | Only GLM 5.3 measured |
| LiveBench research | 73.2% | Not measured | Only GLM 5.3 measured |
| LiveBench browser | 0.0% | Not measured | Only GLM 5.3 measured |
| LiveBench long agentic | 35.9% | Not measured | Only GLM 5.3 measured |
| SREBench (Vals AI) | 9.2% | Not measured | Only GLM 5.3 measured |
| Finance Agent (Vals AI) | 55.8% | 45.4% | GLM 5.3 |
| Tax Agent Bench (Vals AI) | 39.7% | 8.8% | GLM 5.3 |
| Legal Research Bench (Vals AI) | 49.0% | 12.5% | GLM 5.3 |
| Harvey's Legal Agent Benchmark (Vals AI) | 14.6% | Not measured | Only GLM 5.3 measured |
| Excel Modeling Benchmark (Vals AI) | 56.3% | 45.4% | GLM 5.3 |
| Terminal-Bench-Science (Vals AI) | 5.7% | Not measured | Only GLM 5.3 measured |
| τ²-Bench Telecom | Not measured | 83.3% | Only GPT-5.4 Mini measured |
| APEX-Agents | 56.6% | Not measured | Only GLM 5.3 measured |
| DeepResearch Bench | Not measured | 36.3% | Only GPT-5.4 Mini measured |
| Vending-Bench 2 | $8,164 | Not measured | Only GLM 5.3 measured |
| Neither model has a result on the other 23 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index90
| Benchmark | GLM 5.3 | GPT-5.4 Mini | Leader |
|---|---|---|---|
| GPQA Diamond | 88.1% | 83.1% | GLM 5.3 |
| LiveBench reasoning | 85.8% | Not measured | Only GLM 5.3 measured |
| MysteryMechanism (Vals AI) | 23.0% | Not measured | Only GLM 5.3 measured |
| MMMU Pro (Vals AI) | Not measured | 79.3% | Only GPT-5.4 Mini measured |
| GPQA Diamond (Artificial Analysis run) | 91.7% | 87.5% | GLM 5.3 |
| Humanity's Last Exam (Artificial Analysis run) | 42.3% | 28.1% | GLM 5.3 |
| GPQA Diamond (Epoch AI run) | 90.9% | 86.9% | GLM 5.3 |
| ARC-AGI | Not measured | 63.7% | Only GPT-5.4 Mini measured |
| ARC-AGI-2 | Not measured | 18.9% | Only GPT-5.4 Mini measured |
| Chess puzzles | 37.0% | 23.0% | GLM 5.3 |
| Mystery game puzzles | 33.0% | 11.0% | GLM 5.3 |
| LMArena Text: hard prompts | 1490 | 1424 | GLM 5.3 |
| LMArena Text: expert | 1514 | 1435 | GLM 5.3 |
| LMArena Text: hard prompts in English | 1496 | 1420 | GLM 5.3 |
| LMArena Vision: diagrams | Not measured | 1266 | Only GPT-5.4 Mini measured |
| LMArena Vision: homework | Not measured | 1285 | Only GPT-5.4 Mini measured |
| Kagi LLM Benchmark | Not measured | 37.9% | Only GPT-5.4 Mini measured |
| Neither model has a result on the other 5 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index60
| Benchmark | GLM 5.3 | GPT-5.4 Mini | Leader |
|---|---|---|---|
| AIME | Not measured | 95.6% | Only GPT-5.4 Mini measured |
| OTIS Mock AIME | 91.1% | 88.9% | GLM 5.3 |
| FrontierMath, tiers 1 to 3 | 68.8% | 51.2% | GLM 5.3 |
| FrontierMath, tier 4 | 29.3% | 9.8% | GLM 5.3 |
| ProofBench | 49.0% | 21.0% | GLM 5.3 |
| LMArena Text: math | 1495 | 1418 | GLM 5.3 |
| LMArena Text: mathematical work | 1497 | 1429 | GLM 5.3 |
| Neither model has a result on the other 6 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index70
| Benchmark | GLM 5.3 | GPT-5.4 Mini | Leader |
|---|---|---|---|
| MMLU Pro | 86.8% | 84.5% | GLM 5.3 |
| LegalBench (Vals AI) | 84.8% | Not measured | Only GLM 5.3 measured |
| CaseLaw (Vals AI) | Not measured | 51.7% | Only GPT-5.4 Mini measured |
| CorpFin (Vals AI) | Not measured | 60.9% | Only GPT-5.4 Mini measured |
| TaxEval (Vals AI) | 72.4% | 71.2% | GLM 5.3 |
| MortgageTax (Vals AI) | Not measured | 63.5% | Only GPT-5.4 Mini measured |
| MedCode (Vals AI) | 42.9% | Not measured | Only GLM 5.3 measured |
| SimpleQA Verified | 41.0% | 29.4% | GLM 5.3 |
| LMArena Text: science | 1498 | 1417 | GLM 5.3 |
| LMArena Text: medicine and healthcare | 1481 | 1395 | GLM 5.3 |
| LMArena Text: legal and government | 1473 | 1414 | GLM 5.3 |
| LMArena Text: business and finance | 1459 | 1418 | GLM 5.3 |
| Neither model has a result on the other 6 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index9 boards, none shared
| Benchmark | GLM 5.3 | GPT-5.4 Mini | Leader |
|---|---|---|---|
| Design Arena: websites | 1295 | Not measured | Only GLM 5.3 measured |
| Design Arena: UI components | 1314 | Not measured | Only GLM 5.3 measured |
| Design Arena: game dev | 1311 | Not measured | Only GLM 5.3 measured |
| Design Arena: data viz | 1260 | Not measured | Only GLM 5.3 measured |
| Design Arena: 3D | 1332 | Not measured | Only GLM 5.3 measured |
| Design Arena: all code categories | 1304 | Not measured | Only GLM 5.3 measured |
| Design Arena: SVG | 1294 | Not measured | Only GLM 5.3 measured |
| Design Arena: ASCII art | 1208 | Not measured | Only GLM 5.3 measured |
| LiveBench design | 77.3% | Not measured | Only GLM 5.3 measured |
GeneralLike-for-likePublic boards, shown as published, kept out of the index200
| Benchmark | GLM 5.3 | GPT-5.4 Mini | Leader |
|---|---|---|---|
| Vals Index | 59.7% | Not measured | Only GLM 5.3 measured |
| LiveBench data analysis | 70.2% | Not measured | Only GLM 5.3 measured |
| Public Benefits Bench (Vals AI) | 68.5% | Not measured | Only GLM 5.3 measured |
| SAGE (Vals AI) | Not measured | 50.8% | Only GPT-5.4 Mini measured |
| MedScribe (Vals AI) | 88.8% | Not measured | Only GLM 5.3 measured |
| Vals Multimodal Index | Not measured | 54.2% | Only GPT-5.4 Mini measured |
| IFBench | Not measured | 73.3% | Only GPT-5.4 Mini measured |
| Long context reasoning (AA-LCR) | 79.7% | 77.0% | GLM 5.3 |
| DTBench | 87.7% | 80.0% | GLM 5.3 |
| LMCA | 55.5% | 40.8% | GLM 5.3 |
| LMArena Text | 1474 | 1412 | GLM 5.3 |
| LMArena Text: instruction following | 1479 | 1405 | GLM 5.3 |
| LMArena Text: creative writing | 1458 | 1370 | GLM 5.3 |
| LMArena Vision | Not measured | 1245 | Only GPT-5.4 Mini measured |
| LMArena Text: English | 1482 | 1412 | GLM 5.3 |
| LMArena Text: non-English | 1459 | 1405 | GLM 5.3 |
| LMArena Text: Chinese | 1532 | 1446 | GLM 5.3 |
| LMArena Text: German | 1488 | 1408 | GLM 5.3 |
| LMArena Text: Japanese | 1460 | 1375 | GLM 5.3 |
| LMArena Text: Korean | 1460 | 1370 | GLM 5.3 |
| LMArena Text: French | 1503 | 1441 | GLM 5.3 |
| LMArena Text: Spanish | 1462 | 1405 | GLM 5.3 |
| LMArena Text: Russian | 1469 | 1417 | GLM 5.3 |
| LMArena Text: Polish | 1466 | 1415 | GLM 5.3 |
| LMArena Text: multi-turn | 1472 | 1429 | GLM 5.3 |
| LMArena Text: longer queries | 1483 | 1408 | GLM 5.3 |
| LMArena Text: writing and language | 1469 | 1393 | GLM 5.3 |
| LMArena Text: entertainment and media | 1447 | 1373 | GLM 5.3 |
| LMArena Vision: English | Not measured | 1244 | Only GPT-5.4 Mini measured |
| LMArena Vision: Chinese | Not measured | 1286 | Only GPT-5.4 Mini measured |
| LMArena Vision: reading text | Not measured | 1253 | Only GPT-5.4 Mini measured |
| LMArena Vision: humor | Not measured | 1235 | Only GPT-5.4 Mini measured |
| LMArena Vision: recognizing things | Not measured | 1218 | Only GPT-5.4 Mini measured |
| LMArena Vision: creative writing | Not measured | 1222 | Only GPT-5.4 Mini measured |
| Neither model has a result on the other 12 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | GLM 5.3 | GPT-5.4 Mini | 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
GPT-5.4 Mini costs less
Repository review
50K input + 3K output tokens
GPT-5.4 Mini costs less
Agent loop
200K input + 10K output tokens
GPT-5.4 Mini 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 5.3 | OpenAI GPT-5.4 Mini |
|---|---|---|
| Maker | Z.ai | OpenAI |
| Weights | Open weights | No public weights listed |
| Released | August 18, 2026 | March 17, 2026 |
| Context window | 1.05M | 400K |
| Input $/Mtok | $1.40 | $0.75 |
| Output $/Mtok | $4.40 | $4.50 |
| Coding index | 70.6 | 35.8 |
| Independent measurements behind it | 6 | 5 |
| Cost per solved task | $0.36 | $0.70 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is GLM 5.3 better than GPT-5.4 Mini for coding?
Pick GLM 5.3 for coding: it scores 70.6 on the Vibe Coding Index against 35.8 for GPT-5.4 Mini. Pick GPT-5.4 Mini for work measured by LiveCodeBench (Vals AI), where it leads.
Which is cheaper, GLM 5.3 or GPT-5.4 Mini?
GLM 5.3 costs $1.40 per million input tokens and $4.40 per million output tokens; GPT-5.4 Mini costs $0.75 and $4.50. Per solved SWE-bench Verified task, GLM 5.3 costs $0.36 and GPT-5.4 Mini $0.70.
Which has the bigger context window, GLM 5.3 or GPT-5.4 Mini?
GLM 5.3, with 1.05M tokens against 400K.
How many benchmarks is the GLM 5.3 vs GPT-5.4 Mini comparison based on?
66 benchmarks have a result for both models as of October 22, 2026. GLM 5.3's coding index rests on 6 independent benchmarks and GPT-5.4 Mini's on 5. Where only one model has a result, the page says so instead of naming a winner.