OpenAI updated ChatGPT’s default model on May 5, 2026, replacing GPT-5.3 Instant with GPT-5.5 Instant across all users. The update ships with a headline claim that matters more than benchmark scores to business users: 52.5% fewer hallucinated claims on high-stakes prompts covering medicine, law, and finance.

For businesses that have been hesitant to deploy AI in consequential workflows, that number is worth paying attention to.

What Changed

GPT-5.5 Instant is the everyday, speed-optimised model in OpenAI’s lineup — sitting below “Thinking” for complex reasoning and “Pro” for maximum-intensity tasks. It is now the model ChatGPT routes to by default for standard conversations.

The accuracy improvements are substantial:

• 52.5% fewer hallucinated claims compared to GPT-5.3 Instant on high-stakes prompts in medicine, law, and finance
• 37.3% fewer inaccurate claims on conversations that users had previously flagged for factual errors
• AIME 2025 math score of 81.2, up from 65.4 on the previous default
• MMMU-Pro multimodal reasoning score of 76, versus 69.2 previously

Beyond accuracy, the model is faster at deciding when to search the web versus answer from training data — which reduces the kind of confident-but-outdated responses that frustrate users in business settings.

The Memory Feature Businesses Should Notice

The update also rolls out memory source visibility to all ChatGPT models. When GPT-5.5 Instant draws on stored context — past conversations, uploaded files, or linked Gmail data — it now shows users exactly which sources it used and lets them delete or correct stale entries.

This is a meaningful transparency improvement for teams using ChatGPT in professional workflows. If an AI assistant gives advice based on a project brief from six months ago, users can now see that, challenge it, and update the context. The feature is rolling out to Plus and Pro subscribers first, with Business and Enterprise access expected in the coming weeks.

Advanced personalisation that pulls from past chats and files is initially limited to paid tiers. Free users get the improved base model without the personalised memory layer.

What This Means for Business

The 52% hallucination reduction is the number that will matter most to organisations evaluating AI for high-stakes use. A model that fabricates facts in finance, legal, or medical contexts creates real liability. OpenAI’s internal benchmarks suggest this version is meaningfully safer for those categories of work — though real-world use will determine whether the lab’s internal tests hold up under production conditions.

For businesses that have been deploying AI assistants for internal knowledge work, reporting, or customer-facing interactions, the shift to a more accurate default model should improve output quality without any configuration changes. The upgrade happens automatically.

The memory source feature points to a broader direction the enterprise AI space is heading: AI tools that show their reasoning and give users control over the context they draw from. This matters for governance and audit trails — two areas where enterprise adoption has stalled.

From a practical standpoint, teams using ChatGPT for things like contract summaries, financial analysis, or client research should notice fewer instances of the model confidently stating something inaccurate. That does not mean the problem is solved, but a 52% reduction is a material improvement over a previous model that was already widely used in business settings.

The Bigger Picture

OpenAI has shipped seven model updates in the GPT-5 family since the original release in late 2025. The cadence reflects how aggressively the company is iterating now that foundation models are deployed at scale. Each update is an incremental refinement rather than a major capability jump — but the compounding effect across accuracy, speed, and context handling is significant.

For teams that benchmarked AI tools six months ago and held off on deployment, it is worth running those evaluations again. The model that was frustrating in a pilot may behave meaningfully differently today.

Enterprise DNA’s own work with clients across data-intensive industries confirms this pattern: AI tools that felt unreliable in early 2025 pilots have matured considerably. The question for most businesses is no longer whether AI can do the task — it is how to structure the workflow around a tool that keeps getting better.

---

OpenAI’s GPT-5.5 Instant is now the default model for all ChatGPT users. Advanced memory personalisation rolls out to Plus and Pro first, with Business and Enterprise access following in the coming weeks.