Part 3/10:
Up-to-date search integration: The model now includes search functionality, allowing it to fetch current information with web sources directly embedded into responses.
Accessibility for free users: For the first time, free users can access a reasoning model in ChatGPT by selecting the "Reason" mode, democratizing high-level AI functionalities.
Performance and Benchmarks
GPT-3.5 Mini demonstrates remarkable performance improvements in critical areas:
- Scientific and technical reasoning: On the Aim Science Benchmark, 03 Mini with high reasoning effort outperforms GPT-3.01 Mini by approximately 20%, and nearly matches the full GPT-3 model's scores, which reach 96.7% accuracy.