GPT-4.5's multilingual performance is reported to be an improvement over GPT-4. According to OpenAI, GPT-4.5 outperforms GPT-4 on human-translated benchmarks in many languages. This suggests that GPT-4.5 has enhanced capabilities in handling and understanding diverse linguistic nuances across multiple languages[5].
In evaluating multilingual performance, OpenAI translated the MMLU test set into 14 languages using professional human translators. This approach allowed for a more accurate assessment of how well GPT-4.5 can process and respond in different languages compared to GPT-4[4]. The results indicate that GPT-4.5 is more adept at handling multilingual tasks, which is beneficial for applications requiring communication across different languages.
While GPT-4 already demonstrated strong multilingual capabilities, GPT-4.5 further refines these abilities, making it more suitable for global communication and tasks that require understanding and generating text in various languages. However, specific metrics or detailed comparisons of their performance in each language are not provided in the available information.
Overall, GPT-4.5's advancements in multilingual support are part of its broader improvements in language understanding and generation, positioning it as a more versatile tool for diverse linguistic environments.
Citations:
[1] https://openai.com/index/introducing-gpt-4-5/
[2] https://datasciencedojo.com/blog/gpt-3-5-vs-gpt-4-debate/
[3] https://www.theverge.com/news/620021/openai-gpt-4-5-orion-ai-model-release
[4] https://cdn.openai.com/gpt-4-5-system-card.pdf
[5] https://www.reddit.com/r/OpenAI/comments/1iznny5/openai_gpt45_system_card/
[6] https://www.aubergine.co/insights/gpt-3-5-vs-gpt-4-an-in-depth-analysis-of-openais-language-models
[7] https://www.wired.com/story/openai-gpt-45/
[8] https://topmostads.com/gpt-4-5-vs-gpt-5-release/