GPT-5 excels in coding, health, and reasoning tasks

In Short:
– GPT-5 achieves 74.9% on coding tasks and excels in “vibe coding” for software generation.
– It outperforms earlier versions in health queries and scored 94.6% in mathematics.

GPT-5 shows significant advancements in artificial intelligence, reaching 74.9% on SWE-bench Verified and 88% on Aider Polyglot for coding tasks. A

ccording to OpenAI, it excels in “vibe coding,” enabling the generation of complete software from single prompts with minimal guidance.

Initial testing partners like Cursor, Windsurf, and Vercel reported enhancements in code quality and fewer errors than previous versions.

Health Improvements

The model also outperformed earlier versions in health-related queries, achieving 46.2% on HealthBench Hard compared to 31.6% for its predecessor.

OpenAI clarified that GPT-5 should aid users in understanding medical information but is not a substitute for professional medical advice.

In mathematics, GPT-5 scored 94.6% on AIME 2025. Its capabilities extend to surpassing human performance on the SimpleBench, scoring 90% against an average score of 83%.