GPT-5.5vsGPT-5.4
Across 17 shared benchmarks, GPT-5.5 leads overall: GPT-5.5 wins 16, GPT-5.4 wins 1, with 0 ties and an average score difference of +7.69.
GPT-5.516 wins(94%)(6%)1 winGPT-5.4
Benchmark scores
Grouped by capability, sorted by largest gap within each. 17 shared benchmarks.
Math and Reasoning
GPT-5.5 5/5| Benchmark | GPT-5.5 | GPT-5.4 | Diff |
|---|---|---|---|
| MathArena Apex | 80.212 / 17Extra-High (With Tools) | 54.178 / 17Extra-High (With Tools) | +26.04 |
| FrontierMath Tier 4 v2 | 72.507 / 42Extra-High (No Tools) | 4912 / 42Extra-High (No Tools) | +23.50 |
| FrontierMath v2 | 85.266 / 58Extra-High (No Tools) | 78.6010 / 58Extra-High (No Tools) | +6.67 |
| AIME 2026 | 1001 / 29Extra-High (With Tools) | 99.174 / 29Extra-High (With Tools) | +0.83 |
| HMMT Feb 2026 | 98.481 / 10Extra-High (With Tools) | 97.732 / 10Extra-High (With Tools) | +0.75 |
Coding and Software Engineer
GPT-5.5 3/3| Benchmark | GPT-5.5 | GPT-5.4 | Diff |
|---|---|---|---|
| DeepSWE | 67.0429 / 85Extra-High (With Tools) | 51.7758 / 85Extra-High (With Tools) | +15.27 |
| WeirdML v2 | 67.1522 / 52Normal (With Tools) | 57.4431 / 52Normal (With Tools) | +9.71 |
| GSO | 40.204 / 21Extra-High (With Tools) | 31.376 / 21Extra-High (With Tools) | +8.83 |
General Knowledge
GPT-5.5 2/2| Benchmark | GPT-5.5 | GPT-5.4 | Diff |
|---|---|---|---|
| HLE | 13.70351 / 563Normal (No Tools) · Text only | 11.30374 / 563Normal (No Tools) · Text only | +2.40 |
| LiveBench | 79.911 / 117Deep Thinking (No Tools) | 77.975 / 117Deep Thinking (No Tools) | +1.94 |
Agent Level Benchmark
GPT-5.5 1/1| Benchmark | GPT-5.5 | GPT-5.4 | Diff |
|---|---|---|---|
| τ²-Bench - Telecom | 69.30139 / 264Normal (With Tools) | 64.30151 / 264Normal (With Tools) | +5 |
AI Agent - Tool Usage
GPT-5.5 1/1| Benchmark | GPT-5.5 | GPT-5.4 | Diff |
|---|---|---|---|
| MCP-Atlas | 75.3025 / 43Extra-High (With Tools) | 70.6028 / 43Extra-High (With Tools) | +4.70 |
Claw-style Agent Evaluation
GPT-5.4 1/1| Benchmark | GPT-5.5 | GPT-5.4 | Diff |
|---|---|---|---|
| PinchBench v2 | 75.5018 / 45Reported best (effort unspecified) | 75.7017 / 45Reported best (effort unspecified) | -0.20 |
General Evaluation
GPT-5.5 1/1| Benchmark | GPT-5.5 | GPT-5.4 | Diff |
|---|---|---|---|
| GPQA Diamond | 77.27264 / 462Normal (No Tools) | 74.80284 / 462Normal (No Tools) | +2.47 |
Long Context
GPT-5.5 1/1| Benchmark | GPT-5.5 | GPT-5.4 | Diff |
|---|---|---|---|
| AA-LCR | 64122 / 170Normal (No Tools) | 58.30130 / 170Normal (No Tools) | +5.70 |
Text Embedding
GPT-5.5 1/1| Benchmark | GPT-5.5 | GPT-5.4 | Diff |
|---|---|---|---|
| Context Arena | 41.9398 / 126Normal (No Tools) | 32.79112 / 126Normal (No Tools) | +9.14 |
Writing and Creative Capabilities
GPT-5.5 1/1| Benchmark | GPT-5.5 | GPT-5.4 | Diff |
|---|---|---|---|
| Creative Writing | 1,84412 / 106Normal (No Tools) | 1,83614 / 106Normal (No Tools) | +7.90 |
Specs
| Field | GPT-5.5 | GPT-5.4 |
|---|---|---|
| Publisher | OpenAI | OpenAI |
| Release date | 2026-04-23 | 2026-03-05 |
| Model type | Reasoning model | Multimodal model |
| Architecture | Dense | Dense |
| Parameters | Not available | Not available |
| Context length | 1000K | 1M |
| Max output | 128K | 125K |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | GPT-5.5 | GPT-5.4 |
|---|---|---|
| Text input | $5 / 1M tokens | $2.5 / 1M tokens |
| Text output | $30 / 1M tokens | $15 / 1M tokens |
| Cache read | $0.5 / 1M tokens | $0.25 / 1M tokens |
| Cache write | $6.25 / 1M tokens | Not public |
Summary
- GPT-5.5leads in:Math and Reasoning (5/5), Coding and Software Engineer (3/3), General Knowledge (2/2), Agent Level Benchmark (1/1), AI Agent - Tool Usage (1/1), General Evaluation (1/1), Long Context (1/1), Text Embedding (1/1), Writing and Creative Capabilities (1/1)
- GPT-5.4leads in:Claw-style Agent Evaluation (1/1)
On average across the 17 shared benchmarks, GPT-5.5 scores 7.69 higher.
Largest single-benchmark gap: MathArena Apex — GPT-5.5 80.21 vs GPT-5.4 54.17 (+26.04).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.