GPT-5.5 Instant Cuts ChatGPT Hallucinations by 52.5%

OpenAI has upgraded ChatGPT’s default engine to GPT-5.5 Instant. The release replaces GPT-5.3 Instant for hundreds of millions of users, focusing on factual reliability and a deliberate reduction in verbosity. Developers can access the new model immediately via the API using the chat-latest alias.

This release also resolves a widespread anomaly known as the Goblin issue, where previous models occasionally forced references to goblins into unrelated responses.

Benchmarks and Efficiency

Internal evaluations show GPT-5.5 Instant produced 52.5% fewer hallucinated claims than its predecessor on high-stakes prompts in medicine, law, and finance. If you build applications in these sectors, understanding why AI hallucinates remains critical, but the baseline reliability of the default model has shifted. Inaccurate claims dropped by 37.3% on challenging conversation patterns previously flagged by users.

The model also demonstrates significantly stronger reasoning in STEM categories. It scored 81.2 on the AIME 2025 math test, a sharp increase from the previous generation.

Metric	GPT-5.3 Instant	GPT-5.5 Instant
AIME 2025 Score	65.4	81.2
Word Count Reduction	Baseline	-30.2%
Hallucination Rate (High-Stakes)	Baseline	-52.5%

OpenAI tuned the model to be intentionally concise. GPT-5.5 Instant uses 30.2% fewer words and 29.2% fewer lines on average. This eliminates excessive formatting, unnecessary emojis, and unwanted follow-up questions while preserving a conversational tone. It also features updated logic for analyzing image uploads and better decision routing for triggering web searches.

Personalization and Memory Sources

The web interface now supports deeper personalization for Plus and Pro users. The model pulls context from past chats, uploaded files, and connected Gmail accounts. This Gmail integration will expand to mobile, Free, Go, Business, and Enterprise users in the coming weeks.

A new Memory Sources feature addresses the opacity of adding memory to AI agents. The interface now shows users exactly which files or past conversations influenced a specific response. Users can view, edit, or delete these contextual sources directly from the chat window.

Safety and API Availability

GPT-5.5 Instant is the first model in the Instant tier classified as High capability under OpenAI’s Preparedness Framework for Cybersecurity and Biological & Chemical Preparedness. This designation triggered specific safeguards previously restricted to flagship models.

The global rollout to consumer interfaces begins today. GPT-5.3 Instant will remain accessible to paid users for a three-month transition window before it is officially sunset.

If your production systems use the chat-latest endpoint, you are already serving GPT-5.5 Instant. You should monitor your application logs to ensure the 30% reduction in verbosity does not break downstream text parsing, evaluation pipelines, or prompt structures that relied on the longer outputs of previous models.