Overview
OpenAI has released GPT-4.5, an incremental but significant update featuring enhanced reasoning capabilities and more reliable outputs.
Key Improvements
- Better Reasoning: More accurate on complex tasks
- Reduced Hallucinations: Fewer false statements
- Instruction Following: Better adherence to instructions
- Larger Context: 256K token context window
Performance Benchmarks
| Benchmark | GPT-4 | GPT-4.5 |
|---|---|---|
| MMLU | 86.4% | 89.2% |
| HumanEval | 87.1% | 91.4% |
| MATH | 52.9% | 68.3% |
Resources
Source: OpenAI