GPT-5.4vsGPT-5.2
Across 9 shared benchmarks, GPT-5.4 leads overall: GPT-5.4 wins 7, GPT-5.2 wins 2, with 0 ties and an average score difference of +3.61.
GPT-5.47 wins(78%)(22%)2 winsGPT-5.2
Benchmark scores
Grouped by capability, sorted by largest gap within each. 9 shared benchmarks.
General Knowledge
GPT-5.4 3/4| Benchmark | GPT-5.4 | GPT-5.2 | Diff |
|---|---|---|---|
| ARC-AGI-2 | 77.107 / 59Normal (No Tools) | 54.2020 / 59深度思考（无工具、并行） | +22.90 |
| HLE | 52.1014 / 157极高强度思考（工具） | 45.5032 / 157Deep Thinking (With Tools + Internet) | +6.60 |
| ARC-AGI | 93.707 / 65Normal (No Tools) | 90.5015 / 65深度思考（无工具、并行） | +3.20 |
| GPQA Diamond | 92.8010 / 178极高强度思考（无工具） | 93.208 / 178深度思考（无工具、并行） | -0.40 |
Math and Reasoning
GPT-5.4 2/2| Benchmark | GPT-5.4 | GPT-5.2 | Diff |
|---|---|---|---|
| FrontierMath - Tier 4 | 27.1011 / 80极高强度思考（无工具） | 18.8016 / 80Thinking High (No Tools) | +8.30 |
| FrontierMath | 47.605 / 60极高强度思考（无工具） | 40.308 / 60极高强度思考（工具） | +7.30 |
Agent Level Benchmark
GPT-5.2 1/1| Benchmark | GPT-5.4 | GPT-5.2 | Diff |
|---|---|---|---|
| τ²-Bench - Telecom | 64.3030 / 35Normal (With Tools) | 98.704 / 35极高强度思考（工具） | -34.40 |
AI Agent - Information Search
GPT-5.4 1/1| Benchmark | GPT-5.4 | GPT-5.2 | Diff |
|---|---|---|---|
| BrowseComp | 82.7011 / 45极高强度思考（工具） | 65.8024 / 45Deep Thinking (With Tools + Internet) | +16.90 |
Coding and Software Engineer
GPT-5.4 1/1| Benchmark | GPT-5.4 | GPT-5.2 | Diff |
|---|---|---|---|
| SWE-Bench Pro - Public | 57.7010 / 43极高强度思考（无工具） | 55.6017 / 43极高强度思考（工具） | +2.10 |
Specs
| Field | GPT-5.4 | GPT-5.2 |
|---|---|---|
| Publisher | OpenAI | OpenAI |
| Release date | 2026-03-05 | 2025-12-11 |
| Model type | Multimodal model | Chat model |
| Architecture | Dense | Dense |
| Parameters | Not available | Not available |
| Context length | 1M | 400K |
| Max output | 125K | Not available |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | GPT-5.4 | GPT-5.2 |
|---|---|---|
| Text input | $2.5 / 1M tokens | $1.75 / 1M tokens |
| Text output | $15 / 1M tokens | $14 / 1M tokens |
| Cache read | Not public | $0.175 / 1M tokens |
| Cache write | $0.25 / 1M tokens | $1.75 / 1M tokens |
Summary
- GPT-5.4leads in:General Knowledge (3/4), Math and Reasoning (2/2), AI Agent - Information Search (1/1), Coding and Software Engineer (1/1)
- GPT-5.2leads in:Agent Level Benchmark (1/1)
On average across the 9 shared benchmarks, GPT-5.4 scores 3.61 higher.
Largest single-benchmark gap: τ²-Bench - Telecom — GPT-5.4 64.30 vs GPT-5.2 98.70 (-34.40).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.