GPT-5.4vsGemini 3.1 Pro Preview
Across 11 shared benchmarks, GPT-5.4 leads overall: GPT-5.4 wins 6, Gemini 3.1 Pro Preview wins 3, with 2 ties and an average score difference of -0.36.
GPT-5.4
OpenAI · 2026-03-05 · Multimodal model
Gemini 3.1 Pro Preview
Google Deep Mind · 2026-02-20 · Multimodal model
GPT-5.46 wins(55%)Ties2(27%)3 winsGemini 3.1 Pro Preview
Benchmark scores
Grouped by capability, sorted by largest gap within each. 11 shared benchmarks.
General Knowledge
Even 4/4| Benchmark | GPT-5.4 | Gemini 3.1 Pro Preview | Diff |
|---|---|---|---|
| GPQA Diamond | 92.8010 / 178极高强度思考（无工具） | 94.303 / 178Thinking High (No Tools) | -1.50 |
| HLE | 52.1014 / 157极高强度思考（工具） | 51.4015 / 157Thinking High (With Tools) | +0.70 |
| ARC-AGI-3 | 04 / 6Thinking High (No Tools) | 03 / 6Thinking High (No Tools) | — |
| ARC-AGI-2 | 77.107 / 59Normal (No Tools) | 77.107 / 59Thinking High (No Tools) | — |
Math and Reasoning
GPT-5.4 2/2| Benchmark | GPT-5.4 | Gemini 3.1 Pro Preview | Diff |
|---|---|---|---|
| FrontierMath | 47.605 / 60极高强度思考（无工具） | 36.9011 / 60Thinking High (No Tools) | +10.70 |
| FrontierMath - Tier 4 | 27.1011 / 80极高强度思考（无工具） | 16.7020 / 80Normal (No Tools) | +10.40 |
Agent Level Benchmark
Gemini 3.1 Pro Preview 1/1| Benchmark | GPT-5.4 | Gemini 3.1 Pro Preview | Diff |
|---|---|---|---|
| τ²-Bench - Telecom | 64.3030 / 35Normal (With Tools) | 99.301 / 35Thinking High (With Tools) | -35 |
AI Agent - Information Search
Gemini 3.1 Pro Preview 1/1| Benchmark | GPT-5.4 | Gemini 3.1 Pro Preview | Diff |
|---|---|---|---|
| BrowseComp | 82.7011 / 45极高强度思考（工具） | 85.903 / 45Thinking High (With Tools + Internet) | -3.20 |
AI Agent - Tool Usage
GPT-5.4 1/1| Benchmark | GPT-5.4 | Gemini 3.1 Pro Preview | Diff |
|---|---|---|---|
| Terminal Bench 2.0 | 75.104 / 46极高强度思考（工具） | 68.508 / 46Thinking High (With Tools) | +6.60 |
Claw-style Agent Evaluation
GPT-5.4 1/1| Benchmark | GPT-5.4 | Gemini 3.1 Pro Preview | Diff |
|---|---|---|---|
| Pinch Bench | 90.501 / 37Thinking (With Tools) | 86.7010 / 37Thinking (With Tools) | +3.80 |
Coding and Software Engineer
GPT-5.4 1/1| Benchmark | GPT-5.4 | Gemini 3.1 Pro Preview | Diff |
|---|---|---|---|
| SWE-Bench Pro - Public | 57.7010 / 43极高强度思考（无工具） | 54.2023 / 43Thinking High (With Tools) | +3.50 |
Specs
| Field | GPT-5.4 | Gemini 3.1 Pro Preview |
|---|---|---|
| Publisher | OpenAI | Google Deep Mind |
| Release date | 2026-03-05 | 2026-02-20 |
| Model type | Multimodal model | Multimodal model |
| Architecture | Dense | Dense |
| Parameters | Not available | Not available |
| Context length | 1M | 1M |
| Max output | 125K | 32K |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | GPT-5.4 | Gemini 3.1 Pro Preview |
|---|---|---|
| Text input | $2.5 / 1M tokens | $2 / 1M tokens |
| Text output | $15 / 1M tokens | $12 / 1M tokens |
| Cache write | $0.25 / 1M tokens | Not public |
Summary
- GPT-5.4leads in:Math and Reasoning (2/2), AI Agent - Tool Usage (1/1), Claw-style Agent Evaluation (1/1), Coding and Software Engineer (1/1)
- Gemini 3.1 Pro Previewleads in:Agent Level Benchmark (1/1), AI Agent - Information Search (1/1)
- Tied in:General Knowledge
On average across the 11 shared benchmarks, Gemini 3.1 Pro Preview scores 0.36 higher.
Largest single-benchmark gap: τ²-Bench - Telecom — GPT-5.4 64.30 vs Gemini 3.1 Pro Preview 99.30 (-35).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.