Compare / GPT-5.4 vs GLM 5.3
GPT-5.4 vs GLM 5.3
21 results from 5 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | GPT-5.4 | GLM 5.3 |
|---|---|---|---|
| APEX-Agentsreported | Tasks passed% of tasks | 52.4% | 56.6% |
| Arena (formerly LMArena)reported | Confirmed task successIPS effect estimate | -0.04 | 0.06 |
| Arena (formerly LMArena)reported | Praise over complaintIPS effect estimate | -0.01 | 0.06 |
| Arena (formerly LMArena)reported | SteerabilityIPS effect estimate | 0.02 | 0.03 |
| Arena (formerly LMArena)reported | Tool groundingIPS effect estimate | 0.00 | 0.00 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,485 | 1,468 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,482 | 1,474 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,443 | 1,456 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,515 | 1,516 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,471 | 1,480 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,475 | 1,480 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,465 | 1,470 |
| Artificial Analysisreported | Artificial Analysis Coding Indexindex score | 71.1 | 74.8 |
| Artificial Analysisreported | Artificial Analysis Intelligence Indexindex score | 39.0 | 44.8 |
| Artificial Analysisreported | GPQA Diamond% of questions | 92.0% | 91.7% |
| Artificial Analysisreported | Humanity's Last Exam% of questions | 43.7% | 42.3% |
| Artificial Analysisreported | Long-context reasoning (AA-LCR)% of questions | 82.0% | 79.7% |
| Artificial Analysisreported | Terminal-Bench 2.1% of tasks | 78.3% | 83.9% |
| Artificial Analysisreported | Τ-bench banking% of tasks | 39.6% | 50.3% |
| SimpleQA Verified (Epoch AI)reported | Correct answers% of questions | 45.1% | 41.0% |
| Vending-Bench 2reported | Money after a yearUS dollars | $6,144.18 | $8,163.61 |
Bold green marks the better value on that metric. Where a source reports ranges that overlap, the difference may not be meaningful; see the benchmark page for ranges.