← Back to feed

OpenAI promoted GPT-5.5 Instant to the default ChatGPT model on May 5, replacing GPT-5.3 Instant. The change takes effect for all users without requiring any setting changes.

Benchmark Numbers

OpenAI’s internal evals show GPT-5.5 Instant outperforms its predecessor on three dimensions that directly affect daily use:

• 52.5% fewer hallucinated claims on high-stakes medicine, law, and finance prompts
• 37.3% fewer inaccurate claims on conversations users had previously flagged for factual errors
• 30.2% fewer words and 29.2% fewer lines in head-to-head output comparisons — OpenAI is tuning for answers that stop when the explanation is complete

The hallucination reduction is the substantive claim. High-stakes domains — medicine, law, finance — are where confabulation does real damage, and a 52.5% improvement over GPT-5.3 Instant is meaningful if the eval methodology holds under third-party scrutiny. Independent verification is pending.

Personalization Layer

GPT-5.5 Instant pulls context from saved memories, past conversations, uploaded files, and connected Gmail when that context would improve the answer. This is a significant expansion of what “default ChatGPT” means: the model now has persistent context from your history by default, not just within a session.

The Gmail integration is opt-in at the account level, but once connected, GPT-5.5 Instant uses it automatically on relevant queries. This positions ChatGPT closer to a personal assistant than a stateless query interface.

Pricing and Context

GPT-5.5 Instant routes via OpenRouter as openai/gpt-chat-latest at $5/M input and $30/M output tokens, with a 400K context window. Released May 5, 2026.

For comparison: GPT-5.5 (high) — the reasoning-enabled variant — sits at a substantially higher price point. Instant is the volume-serving tier, optimized for consumer ChatGPT load at low latency.

Where This Sits in the Lineup

GPT-5.5 (xhigh) leads the Artificial Analysis Intelligence Index at score 60, followed by GPT-5.5 (high) at 59 and Claude Opus 4.7 at 57. GPT-5.5 Instant is a separate model below these reasoning tiers — it is not the same model as GPT-5.5 (high), and OpenAI has not published its Intelligence Index placement separately.

The practical implication: the version of ChatGPT that hundreds of millions of users interact with daily is now GPT-5.5 Instant. The reasoning benchmarks at the top of the intelligence leaderboard describe a different product.