OpenAI Simply Upgraded ChatGPT’s Default Mannequin—This is What GPT-5.5 On the spot Truly Does

OpenAI simply swapped out the engine inside ChatGPT. Beginning at this time, GPT-5.5 Instant replaces GPT-5.3 On the spot because the default mannequin utilized by tons of of thousands and thousands of people that open ChatGPT each day.

This is not a flashy launch; no new mode, no jaw-dropping demo. However “small enchancment” is a relative time period when the improve cuts hallucinations by greater than half.

What’s GPT-5.5 On the spot?

OpenAI’s GPT household ships in tiers. On the spot is the on a regular basis mannequin, constructed for velocity and common use; Pondering is the slower, extra analytical model for advanced issues; and Professional is the heavyweight for maximum-intensity duties.

GPT-5.5 On the spot is the newest replace to the tier that the majority ChatGPT customers will work together with, whether or not they notice it or not.

In response to OpenAI, the brand new mannequin produced fewer hallucinated claims than GPT-5.3 On the spot on high-stakes prompts in medication, regulation, and finance. Hallucinations have been ChatGPT’s most persistent flaw for the reason that starting.

OpenAI additionally examined in opposition to conversations actual customers had beforehand flagged for factual errors. On these, inaccurate claims dropped by 37.3%.

On HealthBench—a benchmark testing AI responses to actual medical questions, scored from 0 to 100— GPT-5.5 On the spot scores 51.4 factors, up from 49.6. On HealthBench Skilled, the clinical-use model, it jumps from 32.9 to 38.4 factors.

Well being questions are among the many commonest issues folks ask ChatGPT, which makes getting them proper greater than a benchmark train. These outcomes imply GPT 5.5 On the spot elevated accuracy by responding appropriately 38.4% of the time.

GPT-5.5 On the spot additionally pulls extra actively out of your previous chats, saved recordsdata, and linked Gmail account to make solutions personally related. Now when it does this, it exhibits you precisely what context it used, and allows you to delete or appropriate it. “You stay answerable for what’s in your reminiscence,” OpenAI wrote. Non permanent chats nonetheless choose out totally.

The place it suits—and what it is not

When Decrypt coated the GPT-5.5 family launch two weeks in the past, the story was agentic coding and terminal workflows. GPT-5.5 On the spot is a unique animal—it handles extra “fundamental” stuff like your meal plans and electronic mail drafts, not autonomous multi-step coding pipelines. Don’t ask us about GPT 5.4 On the spot, although. It’s most likely chilling subsequent to the O2 mannequin that by no means existed.

The total GPT-5.5 scores 82.7% on Terminal-Bench 2.0, which measures advanced command-line activity efficiency. On the spot is what the remainder of us get, and doubtless what many of the customers will most likely be high-quality working with.

One notable footnote within the system card: GPT-5.5 On the spot is the primary On the spot-tier mannequin OpenAI classifies as “Excessive Functionality” in each cybersecurity and organic domains—succesful sufficient to require the identical automated safeguards beforehand reserved for the extra highly effective Pondering variants. It will not show you how to hack something, however OpenAI constructed guardrails in case somebody tries.

The earlier default, GPT-5.3 On the spot, launched in March with guarantees of fewer preachy refusals and higher accuracy. GPT-5.5 On the spot continues that trajectory. Paid subscribers preferring the outdated model have three months earlier than GPT-5.3 On the spot is retired. Enhanced personalization by way of Gmail rolls out first to Plus and Professional customers on the internet, with Free, Go, Enterprise, and Enterprise to observe within the coming weeks.