GPT-5.4 ProvsGPT-5.2 Pro
Across 6 shared benchmarks, GPT-5.4 Pro leads overall: GPT-5.4 Pro wins 6, GPT-5.2 Pro wins 0, with 0 ties and an average score difference of +10.10.
GPT-5.4 Pro
OpenAI · 2026-03-05 · Multimodal model
GPT-5.2 Pro
OpenAI · 2025-12-11 · Reasoning model
GPT-5.4 Pro6 wins(100%)(0%)0 winsGPT-5.2 Pro
Benchmark scores
Grouped by capability, sorted by largest gap within each. 6 shared benchmarks.
General Knowledge
GPT-5.4 Pro 4/4| Benchmark | GPT-5.4 Pro | GPT-5.2 Pro | Diff |
|---|---|---|---|
| ARC-AGI-2 | 83.308 / 62Thinking High (No Tools) | 54.2023 / 62 | +29.10 |
| HLE | 58.705 / 172Thinking High (With Tools) | 5029 / 172 | +8.70 |
| ARC-AGI | 94.507 / 68Thinking High (No Tools) | 90.5017 / 68 | +4 |
| GPQA Diamond | 94.402 / 187Thinking High (No Tools) | 93.209 / 187 | +1.20 |
AI Agent - Information Search
GPT-5.4 Pro 1/1| Benchmark | GPT-5.4 Pro | GPT-5.2 Pro | Diff |
|---|---|---|---|
| BrowseComp | 89.304 / 53Thinking High (With Tools) | 77.9020 / 53 | +11.40 |
Math and Reasoning
GPT-5.4 Pro 1/1| Benchmark | GPT-5.4 Pro | GPT-5.2 Pro | Diff |
|---|---|---|---|
| FrontierMath - Tier 4 | 37.505 / 80Normal (With Tools + Internet) | 31.309 / 80 | +6.20 |
Specs
| Field | GPT-5.4 Pro | GPT-5.2 Pro |
|---|---|---|
| Publisher | OpenAI | OpenAI |
| Release date | 2026-03-05 | 2025-12-11 |
| Model type | Multimodal model | Reasoning model |
| Architecture | Dense | Dense |
| Parameters | Not available | Not available |
| Context length | 1M | 256K |
| Max output | 125K | Not available |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | GPT-5.4 Pro | GPT-5.2 Pro |
|---|---|---|
| Text input | $30 / 1M tokens | $21 / 1M tokens |
| Text output | $180 / 1M tokens | $168 / 1M tokens |
Summary
- GPT-5.4 Proleads in:General Knowledge (4/4), AI Agent - Information Search (1/1), Math and Reasoning (1/1)
On average across the 6 shared benchmarks, GPT-5.4 Pro scores 10.10 higher.
Largest single-benchmark gap: ARC-AGI-2 — GPT-5.4 Pro 83.30 vs GPT-5.2 Pro 54.20 (+29.10).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.