List your company

OpenAI GPT-5 System Card

About

This is the system card published alongside the OpenAI GPT-5 launch, August 2025. GPT-5 is a unified system with a smart and fast model that answers most questions, a deeper reasoning model for harder problems, and a real-time router that quickly decides which model to use based on conversation type, complexity, tool needs, and explicit intent (for example, if you say 'think hard about this' in the prompt). The router is continuously trained on real signals, including when users switch models, preference rates for responses, and measured correctness, improving over time. Once usage limits are reached, a mini version of each model handles remaining queries. This system card focuses primarily on gpt-5-thinking and gpt-5-main, while evaluations for other models are available in the appendix. The GPT-5 system not only outperforms previous models on benchmarks and answers questions more quickly, but -- more importantly -- is more useful for real-world queries. We've made significant advances in reducing hallucinations, improving instruction following, and minimizing sycophancy, and have leveled up GPT-5's performance in three of ChatGPT's most common uses: writing, coding, and health. All of the GPT-5 models additionally feature safe-completions, our latest approach to safety training to prevent disallowed content. Similarly to ChatGPT agent, we have decided to treat gpt-5-thinking as High capability in the Biological and Chemical domain under our Preparedness Framework, activating the associated safeguards. While we do not have definitive evidence that this model could meaningfully help a novice to create severe biological harm -- our defined threshold for High capability -- we have chosen to take a precautionary approach.

Key insights

What does the GPT-5 system card evaluate?

The card documents GPT-5, a unified system with fast, reasoning, and router models, evaluating safety, hallucinations, deception, and preparedness across biological, cyber, and self-improvement domains.

How does GPT-5 differ from prior models?

The card introduces safe-completions, which aim to provide helpful permitted content instead of only deciding whether to refuse. On evaluated ChatGPT-like prompts, GPT-5-thinking has a reported factual-claim hallucination rate 65% lower than OpenAI o3.

What does this mean for product developers?

The card describes GPT-5 in ChatGPT and the API, with particular attention to writing, coding and health-related queries. Its evaluations help developers understand model behavior and remaining failure modes when choosing integrations.

What limitations and future work do the authors note?

Authors flag regressions in hate/threatening and sexual/exploitative content for gpt-5-main, residual deception in ~2.1% of gpt-5-thinking responses, and evaluation results representing lower bounds on actual model capabilities.

Related benchmarks