GPT-5.4 vs GPT-5.5
GPT-5.5 leads GPT-5.4 by 5.1 points on the Shogo Agent Index, but their ranges overlap, so treat it as close. GPT-5.4 wins 2 and GPT-5.5 wins 20 of 23 shared benchmarks. GPT-5.4 is cheaper for agent workloads.
Add more models to this comparison →Pillars
| Pillar | GPT-5.4 | GPT-5.5 |
|---|---|---|
| Agentic coding | 57 | 63 |
| Long-horizon autonomy | 61 | 67 |
| Knowledge-work agents | 40 | 46 |
| Computer use | 42 | 43 |
| Security & context learning | 54 | 55 |
Shared benchmarks
| Benchmark | GPT-5.4 | GPT-5.5 |
|---|---|---|
| APEX-Agentsindex | 52.4% | 55.1% |
| SWE-bench Verifiedindex | 76.9% | 80.6% |
| Terminal-Bench 2.0index | 81.8% | 84.7% |
| DeepSWEindex | 51.8% | 67.0% |
| Vending-Bench 2index | $6,144 | $7,524 |
| DeepResearch Benchindex | 35.1% | 54.0% |
| GSO-Benchindex | 31.4% | 40.2% |
| PostTrainBenchindex | 17.8% | 23.9% |
| ARC-AGI | 93.7% | 95.0% |
| ARC-AGI-2 | 74.0% | 85.0% |
| Chess Puzzles | 48.0% | 63.0% |
| CritPt | 23.4% | 27.1% |
| DTBench | 94.4% | 96.0% |
| FrontierMath Tier 4 | 49.0% | 72.5% |
| FrontierMath Tiers 1-3 | 78.6% | 85.3% |
| GPQA Diamond | 93.3% | 94.0% |
| LMCA | 52.0% | 54.3% |
| MirrorCode | 23.3% | 23.3% |
| OTIS Mock AIME | 97.8% | 100.0% |
| ProofBench | 56.0% | 50.0% |
| SciCode | 56.6% | 56.1% |
| SimpleQA Verified | 45.1% | 63.0% |
| WebDev Arena | 1463 | 1510 |
Specs and price
| GPT-5.4 | GPT-5.5 | |
|---|---|---|
| Agent Index | 51.6 (#33) | 56.7 (#19) |
| Agent $/M | $1.27 | $2.55 |
| Input / output $/M | $2.50 / $15.00 | $5.00 / $30.00 |
| Context | 1.1M | 1.1M |
| Released | Mar 5, 2026 | Apr 23, 2026 |
| Weights | Closed | Closed |
| API hosts | 4 | 4 |
Don’t pick one. Route.
Shogo can run GPT-5.4 and GPT-5.5 side by side and send each step of an agent to the one that’s cheaper for it.
Try both in Shogo