Grouped by capability, sorted by largest gap within each. 13 shared benchmarks.
| Benchmark | GPT-5.2 | GPT-5 | Diff |
|---|---|---|---|
| ARC-AGI | 90.5015 / 65深度思考（无工具、并行） | 661 / 65 | +84.50 |
| ARC-AGI-2 | 54.2019 / 58深度思考（无工具、并行） | 056 / 58 | +54.20 |
| HLE | 45.5027 / 149Deep Thinking (With Tools + Internet) | 6.30138 / 149 | +39.20 |
| GPQA Diamond | 93.207 / 175深度思考（无工具、并行） | 77.8081 / 175 | +15.40 |
| Benchmark | GPT-5.2 | GPT-5 | Diff |
|---|---|---|---|
| AIME2025 | 1001 / 106极高强度思考（无工具） | 61.9080 / 106 | +38.10 |
| FrontierMath | 40.308 / 60极高强度思考（工具） | 26.3014 / 60Thinking High (With Tools) | +14 |
| FrontierMath - Tier 4 | 18.8016 / 80Thinking High (No Tools) | 12.5029 / 80Thinking High (No Tools) | +6.30 |
| Benchmark | GPT-5.2 | GPT-5 | Diff |
|---|---|---|---|
| τ²-Bench | 8212 / 40极高强度思考（工具） | 8015 / 40thinking + 使用工具 | +2 |
| τ²-Bench - Telecom | 98.704 / 35极高强度思考（工具） | 96.7011 / 35Thinking High (With Tools) | +2 |
| Benchmark | GPT-5.2 | GPT-5 | Diff |
|---|---|---|---|
| SWE-Bench Pro - Public | 55.6012 / 36极高强度思考（工具） | 36.3034 / 36high | +19.30 |
| SWE-bench Verified | 8012 / 103极高强度思考（工具） | 72.8041 / 103high | +7.20 |
| Benchmark | GPT-5.2 | GPT-5 | Diff |
|---|---|---|---|
| BrowseComp | 65.8022 / 43Deep Thinking (With Tools + Internet) | 54.9030 / 43thinking + 使用工具 | +10.90 |
| Benchmark | GPT-5.2 | GPT-5 | Diff |
|---|---|---|---|
| MMMU | 85.901 / 28极高强度思考（无工具） | 84.205 / 28high | +1.70 |
| Field | GPT-5.2 | GPT-5 |
|---|---|---|
| Publisher | OpenAI | OpenAI |
| Release date | 2025-12-11 | 2025-08-07 |
| Model type | AI model | Foundation model |
| Architecture | Dense | Dense |
| Parameters | 0.0 | 0.0 |
| Context length | 400K | 400K |
| Max output | Not available | 131072 |
Prices use DataLearner records when available; missing fields are not inferred.
| Item | GPT-5.2 | GPT-5 |
|---|---|---|
| Text input | $1.75 / 1M tokens | 1.25 美元/100 万tokens |
| Text output | $14 / 1M tokens | 10 美元/100 万tokens |
| Cache read | $0.175 / 1M tokens | Not public |
| Cache write | $1.75 / 1M tokens | Not public |
On average across the 13 shared benchmarks, GPT-5.2 scores 22.68 higher.
Largest single-benchmark gap: ARC-AGI — GPT-5.2 90.50 vs GPT-5 6 (+84.50).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.
