OpenAI GPT-5.3 Instant less likely to beat around the bush

"We heard feedback that GPT‑5.2 Instant would sometimes refuse questions it should be able to answer safely, or respond in ways that feel overly cautious or preachy, particularly around sensitive topics. GPT‑5.3 Instant significantly reduces unnecessary refusals, while toning down overly defensive or moralizing preambles before answering the question."

"On the higher-stakes evaluation, GPT‑5.3 Instant reduces hallucination rates by 26.8 percent when using the web and 19.7 percent when relying only on its internal knowledge, compared to prior models. On the user-feedback evaluation, hallucinations decrease by 22.5 percent with web use and 9.6 percent without web access."

OpenAI released GPT-5.3 Instant, addressing feedback that GPT-5.2 Instant refused questions it could safely answer and responded with overly cautious or preachy language. The new model provides direct answers without unnecessary caveats while maintaining safety. GPT-5.3 Instant significantly reduces hallucinations across multiple evaluation domains: 26.8 percent reduction on high-stakes decisions using web search, 19.7 percent without web access, and 22.5 percent on user-feedback evaluations with web use. The model improves at contextualizing web search results and writing quality. However, performance declined slightly on OpenAI's internal benchmark measurements.

Read at Theregister