OpenAI Just Upgraded ChatGPT's Default Model—Here's What GPT-5.5 Instant Actually Does

OpenAI conscionable swapped retired the motor wrong ChatGPT. Starting today, GPT-5.5 Instant replaces GPT-5.3 Instant arsenic the default exemplary utilized by hundreds of millions of radical who unfastened ChatGPT each day.

This isn't a flashy launch; nary caller mode, nary jaw-dropping demo. But "small improvement" is simply a comparative word erstwhile the upgrade cuts hallucinations by much than half.

What is GPT-5.5 Instant?

OpenAI's GPT household ships successful tiers. Instant is the mundane model, built for velocity and wide use; Thinking is the slower, much analytical mentation for analyzable problems; and Pro is the heavyweight for maximum-intensity tasks.

GPT-5.5 Instant is the latest update to the tier that astir ChatGPT users volition interact with, whether they recognize it oregon not.

According to OpenAI, the caller exemplary produced less hallucinated claims than GPT-5.3 Instant connected high-stakes prompts successful medicine, law, and finance. Hallucinations person been ChatGPT's astir persistent flaw since the beginning.

OpenAI besides tested against conversations existent users had antecedently flagged for factual errors. On those, inaccurate claims dropped by 37.3%.

On HealthBench—a benchmark investigating AI responses to existent aesculapian questions, scored from 0 to 100— GPT-5.5 Instant scores 51.4 points, up from 49.6. On HealthBench Professional, the clinical-use version, it jumps from 32.9 to 38.4 points.

Health questions are among the astir communal things radical inquire ChatGPT, which makes getting them close much than a benchmark exercise. These results mean GPT 5.5 Instant accrued accuracy by responding correctly 38.4% of the time.

GPT-5.5 Instant besides pulls much actively from your past chats, saved files, and connected Gmail relationship to marque answers personally relevant. Now erstwhile it does this, it shows you precisely what discourse it used, and lets you delete oregon close it. "You stay successful power of what's successful your memory," OpenAI wrote. Temporary chats inactive opt retired entirely.

Where it fits—and what it isn't

When Decrypt covered the GPT-5.5 household launch 2 weeks ago, the communicative was agentic coding and terminal workflows. GPT-5.5 Instant is simply a antithetic animal—it handles much “basic” worldly similar your repast plans and email drafts, not autonomous multi-step coding pipelines. Don’t inquire america astir GPT 5.4 Instant, though. It’s astir apt chilling adjacent to the O2 exemplary that ne'er existed.

The afloat GPT-5.5 scores 82.7% connected Terminal-Bench 2.0, which measures analyzable command-line task performance. Instant is what the remainder of america get, and astir apt what astir of the users volition astir apt beryllium good moving with.

One notable footnote successful the strategy card: GPT-5.5 Instant is the archetypal Instant-tier exemplary OpenAI classifies arsenic "High Capability" successful some cybersecurity and biologic domains—capable capable to necessitate the aforesaid automated safeguards antecedently reserved for the much almighty Thinking variants. It won't assistance you hack anything, but OpenAI built guardrails successful lawsuit idiosyncratic tries.

The erstwhile default, GPT-5.3 Instant, launched successful March with promises of less preachy refusals and amended accuracy. GPT-5.5 Instant continues that trajectory. Paid subscribers who similar the aged mentation person 3 months earlier GPT-5.3 Instant is retired. Enhanced personalization via Gmail rolls retired archetypal to Plus and Pro users connected the web, with Free, Go, Business, and Enterprise to travel successful the coming weeks.