GPT-5.5 Instant

Summary

GPT-5.5 Instant is OpenAI's new default ChatGPT model, released May 5, 2026, replacing GPT-5.3 Instant with 52.5% fewer hallucinations and ~30% more concise responses. It introduces personalization across past chats, files, and connected Gmail at roughly 80–85% of standard GPT-5.5's quality.

Overview

GPT-5.5 Instant is OpenAI's latest "Instant" tier of the GPT-5.5 family — designed to be the default ChatGPT model that handles the majority of consumer and free-tier traffic. It was released on May 5, 2026 (the "5/5" date is referenced in OpenAI's launch blog), replacing GPT-5.3 Instant as the ChatGPT default and shipping simultaneously in the API as gpt-5.5-instant (also surfaced as chat-latest). Compared with GPT-5.3 Instant, GPT-5.5 Instant focuses on three things: (1) measurable reductions in hallucination on high-stakes prompts, (2) more concise, less emoji-laden, less sycophantic responses, and (3) personalization grounded in past chats, files, and (for Plus/Pro web users) connected Gmail.

GPT-5.5 Instant is a smaller, faster, and cheaper sibling of standard GPT-5.5 — OpenAI cites quality at roughly 80–85% of standard GPT-5.5 across most benchmarks, with substantially lower latency and cost. It is multimodal (text + image input).

Specifications

  • Developer: OpenAI
  • Release Date: May 5, 2026
  • Type: Large Language Model (LLM); multimodal (text + image input); fast/instant tier
  • Family: GPT-5.5 series
  • Knowledge Cutoff: August 2025
  • Context Window: Inherits the GPT-5.5 family API context (1M tokens; ChatGPT surfaces a smaller working context)
  • API Identifier: gpt-5.5-instant (and chat-latest)
  • Access:
    • Default ChatGPT model for Free, Plus, Pro, Go, Business, and Enterprise (rolling out)
    • API: gpt-5.5-instant / chat-latest
    • Replaces GPT-5.3 Instant; GPT-5.3 Instant remains available to paid users for three months before retirement

Capabilities

Reduced Hallucinations: 52.5% fewer hallucinated claims than GPT-5.3 Instant on high-stakes prompts (medicine, law, finance). 37.3% reduction in inaccurate claims on flagged conversations.

Concision: Roughly 30.2% fewer words and 29.2% fewer lines used to convey the same point as GPT-5.3 Instant. OpenAI specifically targeted "gratuitous emojis" and unnecessary follow-up questions.

Math/Reasoning: 81.2 on AIME 2025 vs. 65.4 for GPT-5.3 Instant — a meaningful jump for the Instant tier.

Multimodal: Text and image input.

Personalization (new):

  • Searches across past chats, uploaded files, and connected Gmail (Plus/Pro on web at launch; mobile + Free/Go/Business/Enterprise rolling out in subsequent weeks)
  • Introduces "memory sources" — UI showing which prior context drove personalization
  • Aimed at making ChatGPT feel meaningfully aware of the user across sessions without dedicated prompt engineering

Quality vs. Standard GPT-5.5: ~80–85% of standard GPT-5.5 quality on most benchmarks, at much lower latency and cost.

Limitations

GPT-5.5 Instant is not OpenAI's most capable model — for the hardest reasoning, agentic-workflow, or 1M-context tasks, OpenAI still recommends standard GPT-5.5 or GPT-5.5 Pro. Personalization features rely on the user's past ChatGPT history, files, and Gmail; users without that context will not see the same uplift. Although the August 2025 knowledge cutoff is recent, it precedes the most consequential AI lab events of 2026 (e.g., Anthropic's Mythos, Meta's Muse Spark, GPT-5.5 itself). Personalization across files and Gmail introduces obvious privacy considerations that OpenAI has addressed via the new "memory sources" UI.

Recent Developments

  • Default ChatGPT Model (May 5, 2026): GPT-5.5 Instant became the default ChatGPT model across tiers, replacing GPT-5.3 Instant. Rollout began on the web and is expanding to mobile and remaining tiers (Free, Go, Business, Enterprise) in subsequent weeks.
  • Hallucination Reductions: 52.5% fewer hallucinated claims on high-stakes prompts and 37.3% reduction in inaccurate claims on flagged conversations versus GPT-5.3 Instant.
  • Concision Push: OpenAI's product team explicitly targeted "gratuitous emojis," unnecessary follow-up questions, and verbose responses, reducing average word count by ~30%.
  • Personalization Launch: New ability to draw on past ChatGPT conversations, files, and (for Plus/Pro on web at launch) connected Gmail when generating responses, surfaced via the new "memory sources" UI.
  • GPT-5.3 Instant Sunset Window: Paid users retain GPT-5.3 Instant access via model picker for three months before retirement.

Last Updated

May 10, 2026

→ Back to Models