See key specs and per-benchmark scores for each model/mode. Scroll horizontally for all columns. 当前对比 2 个模型的评测数据与核心参数。
Gemini 3.1 Pro Preview
Google Deep Mind
GPT-5.4
OpenAI
Each axis is a category average, normalized to a 100-point radar.
Relative edge: Agent能力评测 +17.7 / Relative gap: 编程与软件工程 -21.8
Relative edge: 编程与软件工程 +21.8 / Relative gap: Agent能力评测 -17.7
Method: for each model and benchmark, the chart first averages all scores in the current mode scope instead of taking the best score, then averages those benchmark scores within each category. Only benchmarks with at least two selected models scored are included; missing values are not counted as zero.
Best overall
GPT-5.4 · 69.96
Best single
Gemini 3.1 Pro Preview · τ²-Bench - Telecom 99.30
Modality coverage
Gemini 3.1 Pro Preview · 4 modalities
Head to head
14
Benchmarks
5
Wins
8
Losses
-4.44
Average diff
Compare benchmark results across thinking modes and tool usage.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
Complete scores for each model/mode across selected benchmarks.
14 benchmarks with comparable scores. Each model shows its best score; mode label is displayed below.
| Benchmark | Gemini 3.1 Pro Preview | GPT-5.4 |
|---|---|---|
ARC-AGI-2 综合评估 | 77.10Thinking Level · High | 77.10Standard Mode |
GPQA Diamond 综合评估 | 94.30Thinking Level · High | 92.80Thinking Level · Extra High |
HLE 综合评估 | 51.40Thinking Level · High ｜ Tools | 52.10Thinking Level · Extra High ｜ Tools |
LiveBench 综合评估 | 79.93Thinking Level · High | 80.28Deep Thinking Mode |
DeepSWE 编程与软件工程 | 12.00Thinking Level · High ｜ Tools | 52.00Thinking Level · Extra High ｜ Tools |
SWE-Bench Pro - Public 编程与软件工程 | 54.20Thinking Level · High ｜ Tools | 57.70Thinking Level · Extra High |
τ²-Bench - Telecom Agent能力评测 | 99.30Thinking Level · High ｜ Tools | 98.90Thinking Level · Extra High ｜ Tools |
FrontierMath 数学推理 | 36.90Thinking Level · High | 47.60Thinking Level · Extra High |
16.70Standard Mode | 27.10Thinking Level · Extra High | |
BrowseComp AI Agent - 信息收集 | 85.90Thinking Level · High ｜ Tools | 82.70Thinking Level · Extra High ｜ Tools |
MCP-Atlas AI Agent - 工具使用 | 78.20Thinking Level · High ｜ Tools | 70.60Thinking Level · Extra High ｜ Tools |
OSWorld-Verified AI Agent - 工具使用 | 76.20Thinking Enabled ｜ Tools | 75.00Thinking Level · Extra High ｜ Tools |
Terminal Bench 2.0 AI Agent - 工具使用 | 68.50Thinking Level · High ｜ Tools | 75.10Thinking Level · Extra High ｜ Tools |
Pinch Bench OpenClaw智能体能力综合测评 | 86.70Thinking Enabled ｜ Tools | 90.50Thinking Enabled ｜ Tools |
Side-by-side input/output token pricing
Licensing, MoE architecture, and multi-modality support.
| Features & specs | Gemini 3.1 Pro PreviewGoogle Deep Mind | GPT-5.4OpenAI |
|---|---|---|
Core specsRelease | 2026-02-20 | 2026-03-05 |
Context length | 1M | 1M |
Max output | 65536 | 128000 |
MoE | No | No |
LicenseCode Open Source | Not provided | Not provided |
Weights Open Source | Not provided | Not provided |
Commercial use | 不开源 | 不开源 |
Modality supportText Input/Output | / | / |
Image Input/Output | / | / |
Audio Input/Output | / | Not provided |
Video Input/Output | / | Not provided |
ResourcesPaper / report | Gemini 3.1 Pro: A smarter model for your most complex tasks | Introducing GPT‑5.4 |
