OpenAI launches GPT-5 Chatbot AI version: what’s new?

OpenAI has officially unveiled GPT-5, the company’s first “unified” model that combines fast reaction time with a deep thinking mode. It will replace GPT-4 in ChatGPT and can handle both complex technical tasks (app development, research preparation) and everyday tasks like calendar management.

The key features are switching between modes and reducing the frequency of “hallucinations” to 4.8% in thinking mode (versus 20-22% in its predecessors).

Test results

• SWE-bench Verified (real GitHub problems): 74.9% on the first try, slightly ahead of Claude Opus 4.1 (74.5%).
• GPQA Diamond (PhD-level questions): 89.4% — higher than Claude Opus 4.1 (80.9%) and Grok 4 Heavy (88.9%).
• HealthBench Hard Hallucinations (accuracy in medical topics): 1.6% hallucinations vs. 12.9% for GPT-4o.
• Mixed results in Tau-bench: better on retail sites, but slightly weaker on airline sites.

New ChatGPT Features

Four new character styles for responses: Cynic, Robot, Listener, and Boring. GPT-5 is available to all users, but Plus ($20) subscribers have increased limits, while Pro ($200) has no limits. Corporate plans will get GPT-5 next week.

API access: three versions – gpt-5, gpt-5-mini and gpt-5-nano, cost from $1.25 per million input tokens.