OpenAI GPT-5 Review: Built to Win Benchmarks, Not Hearts

Key Points

1. GPT-5 performs exceptionally well in coding tasks and structured problem-solving, achieving high scores on benchmarks like FrontierMath.

2. The model struggles with creative writing, producing technically correct but emotionally flat narratives compared to competitors like Claude 4.1 Opus.

3. GPT-5 fails at basic arithmetic, such as solving simple equations, despite excelling in advanced mathematical problems.

4. The model is cost-effective compared to competitors like Claude 4.1 Opus, offering strong performance for professional use cases at a lower price.

5. GPT-5 has a context window of up to 400K tokens, which is an improvement over previous models but still lags behind competitors like Gemini and Llama 4 Scout.