GPT-5.4 vs GPT-5.5
Respan Index by category
Each category is scored 0 to 100, and how far ahead the winner is. Leads under a point are close to even.
| Category | GPT-5.4 | GPT-5.5 |
|---|---|---|
| Coding | 25.9 GPT-5.4 | 30.2 GPT-5.5 |
| Agents | 62.0 GPT-5.4 | 64.2 GPT-5.5 |
| Reasoning | 61.4 GPT-5.4 | 68.4 GPT-5.5 |
| Knowledge & language | 46.9 GPT-5.4 | 63.7 GPT-5.5 |
| Human preference | 53.1 GPT-5.4 | 54.4 GPT-5.5 |
Benchmarks both report
GPT-5.5 leads on 34 of the 39 benchmarks both report. Higher is better. Each score links to where it was published.
| Benchmark | GPT-5.4 | GPT-5.5 |
|---|---|---|
| LMArena Elo | 1475 | 1481 |
| LMArena WebDev (LMArena) | 1462 | 1513 |
| MirrorCode (Epoch AI) | 23.3% | 23.3% |
| LMArena Vision (LMArena) | 1282 | 1285 |
| Chess Puzzles (Epoch AI) | 48% | 63% |
| LMArena Agent (LMArena) | 0.0267 | 0.0471 |
| APEX-Agents (Mercor) | 52.4% | 55.1% |
| DeepSWE v1.1 (Datacurve) | 51.8% | 67% |
| EBR-bench (Epoch AI) | 25.4% | 34.3% |
| FrontierMath Tier 4 v2 (Epoch AI) | 49% | 72.5% |
| FrontierMath Tiers 1-3 v2 (Epoch AI) | 78.6% | 85.3% |
| Furniture Assembly (Epoch AI) | 37.5% | 44.2% |
| GPQA Diamond | 92.8% | 90.7% |
| LiveBench Agentic Coding (LiveBench) | 53.8% | 54% |
| LiveBench Coding (LiveBench) | 77.5% | 82.1% |
| LiveBench Data Analysis (LiveBench) | 79.3% | 81.6% |
| LiveBench Instruction Following (LiveBench) | 70.2% | 70.7% |
| LiveBench Language (LiveBench) | 82.6% | 87.4% |
| LiveBench Mathematics (LiveBench) | 94.1% | 95.9% |
| LiveBench Reasoning (LiveBench) | 88.1% | 89.7% |
| Mystery Game Puzzles (Epoch AI) | 37% | 56% |
| OTIS Mock AIME 2024-2025 (Epoch AI) | 97.8% | 84.4% |
| SAGE (Vals AI) | 43.3% | 51.5% |
| SimpleQA Verified | 45.1% | 63% |
| Vending-Bench 2 (Andon Labs) | 6144.18 | 7523.84 |
| WeirdML (Håvard Tveit Ihle) | 77.7% | 84.9% |
| AIME 2026 | 99.2% | 100% |
| tau2-bench Banking Knowledge (Sierra) | 39.4% | 44.6% |
| GSO Opt@1 (GSO) | 30.4% | 37.3% |
| BrowseComp | 82.7% | 84.4% |
| GDPval (wins or ties) | 83% | 84.9% |
| MCP Atlas | 67.2% | 75.3% |
| MMMU-Pro (no tools) | 81.2% | 81.2% |
| MMMU-Pro (with tools) | 82.1% | 83.2% |
| OSWorld-Verified | 75% | 78.7% |
| SWE-Bench Pro | 57.7% | 58.6% |
| Tau2-bench Telecom | 98.9% | 98% |
| Terminal-Bench 2.0 | 75.1% | 82.7% |
| Toolathlon | 54.6% | 55.6% |
FAQ
Is GPT-5.4 better than GPT-5.5?
GPT-5.5 scores 55.2 on the Respan Index (#29), ahead of GPT-5.4 at 50.0 (#43). GPT-5.5 leads on 34 of the 39 benchmarks both report.
Which is cheaper, GPT-5.4 or GPT-5.5?
GPT-5.4 lists at $2.5 input and $15 output per 1M tokens; GPT-5.5 at $5 and $30. GPT-5.4 is 2.0x cheaper per token at list price.
Which has the larger context window?
GPT-5.4 accepts 1.1M tokens and GPT-5.5 1M tokens.