OpenAI just launched GPT-5. Here is what actually changed.

OpenAI has officially released GPT-5, and the AI community is buzzing with reactions ranging from awe to skepticism. After spending a week with the new model, we can separate the signal from the noise.

The benchmarks

On standard reasoning benchmarks, GPT-5 shows a 23% improvement over GPT-4 Turbo. But raw numbers only tell part of the story. Where GPT-5 truly shines is in multi-step reasoning tasks that require holding context across long conversations.

We ran a battery of tests across coding, analysis, and creative writing tasks. The results were consistent: GPT-5 makes fewer logical errors and maintains coherence over much longer interactions.

The pricing question

At $30 per million input tokens and $60 per million output tokens, GPT-5 is roughly 2x the cost of GPT-4 Turbo. For enterprise customers, OpenAI is offering volume discounts that bring it closer to parity, but for individual developers, the cost increase is significant.

The real question is not whether GPT-5 is better. It is whether it is enough better to justify the price.

What matters for developers

The 200K context window is now standard, and the model handles structured output with near-perfect reliability. Function calling has been completely rearchitected, making it far more predictable for production applications.

If you are building AI-powered products, GPT-5 is a meaningful upgrade. If you are using AI for casual tasks, GPT-4 Turbo remains the better value.