GPT-4.5
Multimodal
Zero-eval
#1CharXiv-D
#1SWE-Lancer
#2Graphwalks parents <128k
+7 more
by OpenAI
+
+
+
+
About
GPT-4.5 was developed as an intermediate advancement between GPT-4 and GPT-5, designed to incorporate incremental improvements in reasoning, factuality, and instruction-following. Built to bridge generational capabilities while maintaining compatibility with existing applications, it represents a refinement of the GPT-4 architecture with enhanced performance characteristics.
+
+
+
+
Pricing Range
Input (per 1M)$75.00 -$75.00
Output (per 1M)$150.00 -$150.00
Providers1
+
+
+
+
Timeline
AnnouncedFeb 27, 2025
ReleasedFeb 27, 2025
+
+
+
+
Specifications
Capabilities
Multimodal
+
+
+
+
License & Family
License
Proprietary
Performance Overview
Performance metrics and category breakdown
Overall Performance
27 benchmarks
Average Score
63.1%
Best Score
97.0%
High Performers (80%+)
6Performance Metrics
Max Context Window
132.1KAvg Throughput
50.0 tok/sAvg Latency
20ms+
+
+
+
All Benchmark Results for GPT-4.5
Complete list of benchmark scores with detailed information
| GSM8k | text | 0.97 | 97.0% | Self-reported | |
| MMLU | text | 0.91 | 90.8% | Self-reported | |
| CharXiv-D | multimodal | 0.90 | 90.0% | Self-reported | |
| IFEval | text | 0.88 | 88.2% | Self-reported | |
| HumanEval | text | 0.88 | 88.0% | Self-reported | |
| MMMLU | text | 0.85 | 85.1% | Self-reported | |
| MMMU | multimodal | 0.75 | 75.2% | Self-reported | |
| Graphwalks parents <128k | text | 0.73 | 72.6% | Self-reported | |
| MathVista | multimodal | 0.72 | 72.3% | Self-reported | |
| COLLIE | text | 0.72 | 72.3% | Self-reported |
Showing 1 to 10 of 27 benchmarks
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+