Gemini 3 Pro
Multimodal
#1MMMLU
#1Ï„-bench
#1MMMU-Pro
+2 more
by Google DeepMind
+
+
+
+
About
Gemini 3 Pro, released by Google in November 2025, is a large language model from the Gemini 3 family designed for complex reasoning, broad world knowledge, and multi-step problem solving. It features a 1M token context window, 64K maximum output tokens, and native multimodal input including text, images, audio, and video. Gemini 3 Pro targets research-grade reasoning, advanced data analysis, and applications requiring comprehensive understanding across diverse input formats.
+
+
+
+
Pricing Range
Input (per 1M)$2.50 -$2.50
Output (per 1M)$20.00 -$20.00
Providers2
+
+
+
+
Timeline
ReleasedNov 18, 2025
+
+
+
+
Specifications
Capabilities
Multimodal
+
+
+
+
License & Family
License
Proprietary
Performance Overview
Performance metrics and category breakdown
Overall Performance
17 benchmarks
Average Score
64.6%
Best Score
98.0%
High Performers (80%+)
6Performance Metrics
Max Context Window
1.0MTop Categories
Science
91.9%
Knowledge
91.8%
Multimodal
81.0%
Agents
71.4%
Coding
60.3%
+
+
+
+
All Benchmark Results for Gemini 3 Pro
Complete list of benchmark scores with detailed information
| TAU2-Bench Telecom | Agents | 98.00 | 98.0% | Self-reported | |
| GPQA Diamond | Science | 91.90 | 91.9% | Unverified | |
| MMMLU | Knowledge | 91.80 | 91.8% | Self-reported | |
| Ï„-bench | Agents | 85.40 | 85.4% | Unverified | |
| TAU2-Bench Retail | Agents | 85.30 | 85.3% | Self-reported | |
| MMMU-Pro | Multimodal | 81.00 | 81.0% | Self-reported | |
| SWE Bench Verified | Coding | 78.00 | 78.0% | Unverified | |
| GDPVal AA ELO | Agents | 1201.00 | 60.1% | Self-reported | |
| BrowseComp | Agents | 59.20 | 59.2% | Unverified | |
| Terminal Bench 2.0 | Coding | 56.20 | 56.2% | Unverified |
Showing 1 to 10 of 17 benchmarks