Compare / GPT-5.4 vs GPT-5.5
GPT-5.4 vs GPT-5.5
32 results from 6 sources that measured both models. Each row is in the source's own units; there is no overall winner.
| Benchmark | Metric | GPT-5.4 | GPT-5.5 |
|---|---|---|---|
| Arena (formerly LMArena)reported | Confirmed task successIPS effect estimate | -0.04 | -0.03 |
| Arena (formerly LMArena)reported | Praise over complaintIPS effect estimate | -0.01 | 0.08 |
| Arena (formerly LMArena)reported | SteerabilityIPS effect estimate | 0.02 | 0.07 |
| Arena (formerly LMArena)reported | Tool groundingIPS effect estimate | 0.00 | 0.00 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,470 | 1,483 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,195 | 1,224 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,485 | 1,486 |
| Arena (formerly LMArena)reported | Business, management and financeArena rating | 1,482 | 1,489 |
| Arena (formerly LMArena)reported | Creative writingArena rating | 1,443 | 1,452 |
| Arena (formerly LMArena)reported | Expert promptsArena rating | 1,515 | 1,513 |
| Arena (formerly LMArena)reported | Instruction followingArena rating | 1,471 | 1,478 |
| Arena (formerly LMArena)reported | OverallArena rating | 1,475 | 1,481 |
| Arena (formerly LMArena)reported | Writing, literature and languageArena rating | 1,465 | 1,470 |
| Microsoft STATE-Benchreported | Consistency (pass^5)% of tasks | 41.8% | 44.2% |
| Microsoft STATE-Benchreported | Customer support task success (pass@1)% of tasks | 59.0% | 59.0% |
| Microsoft STATE-Benchreported | Shopping assistant task success (pass@1)% of tasks | 55.0% | 55.0% |
| Microsoft STATE-Benchreported | Task success (pass@1)% of tasks | 58.6% | 58.9% |
| Microsoft STATE-Benchreported | Travel task success (pass@1)% of tasks | 62.0% | 62.0% |
| Microsoft STATE-Benchreported | User experiencescore from 1 to 5 | 3.67 | 3.62 |
| OpenHands Indexreported | Average% resolved | 64.3% | 65.9% |
| OpenHands Indexreported | Front end (SWE-Bench Multimodal)% resolved | 36.8% | 38.2% |
| OpenHands Indexreported | Greenfield (Commit0)% resolved | 56.2% | 43.8% |
| OpenHands Indexreported | Information gathering (GAIA)% resolved | 82.4% | 86.1% |
| OpenHands Indexreported | Issue resolution (SWE-Bench)% resolved | 75.6% | 78.2% |
| OpenHands Indexreported | Testing (SWT-Bench)% resolved | 70.4% | 83.4% |
| tau2-benchreported | Consistency (pass^4)% of tasks | 21.6% | 29.9% |
| tau2-benchreported | Task success (pass^1)% of tasks | 39.4% | 44.6% |
| UGI Leaderboardreported | Requested-length error% off the requested word count | 30.0% | 13.0% |
| UGI Leaderboardreported | Style adherencescore from 0 to 1 | 0.40 | 0.41 |
| UGI Leaderboardreported | Writing scorescore out of 100 | 68.8 | 69.7 |
| Vectara Hallucination Leaderboardreported | Answer rate% of documents | 99.9% | 100.0% |
| Vectara Hallucination Leaderboardreported | Hallucination rate% of summaries | 7.0% | 9.3% |
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.