GPT-5.5 Instant is the new default ChatGPT model optimized for low latency, higher factuality, and more personalized responses leveraging memory from past chats and connected apps.
Release date
5 May 2026
Access:APIHostedDeployment:โ Cloud
Overview
Access & deployment
APIHosted
Cloud
Weights: Closed
Key parameters
๐ฅ Input: text, image
Technical specification
Modalities
โฌ Input
textimage
โฌ Output
textcodestructured_data
Capabilities and applications
Native model capabilities
Reasoning
Category: reasoning
Multimodal understanding
Category: multimodal
Image understanding
Category: vision
Coding
Category: coding
Multilingual
Category: language
Benchmark results
4 benchmarks
Internal hallucination eval (high-stakes prompts)
reduction in hallucinated claims vs GPT-5.3 Instant ยท OpenAI internal evaluations on high-stakes prompts (medicine, law, finance).
52.5%
๐
5 May 2026๐ OpenAI (openai.com/index/gpt-5-5-instant/)
OpenAI internal benchmark; no public methodology available.
Internal factual error eval (user-flagged conversations)
reduction in inaccurate claims vs GPT-5.3 Instant ยท OpenAI internal evaluation on challenging conversations flagged by users for factual errors.
37.3%
๐
5 May 2026๐ OpenAI (openai.com/index/gpt-5-5-instant/)
OpenAI internal benchmark; no public methodology available.
Response conciseness โ words
fewer words vs GPT-5.3 Instant ยท Response length comparison (word count) on advice-style sample prompts.
30.2%
๐
5 May 2026๐ OpenAI (openai.com/index/gpt-5-5-instant/)
Result from OpenAI blog post example; full methodology not disclosed.
Response conciseness โ lines
fewer lines vs GPT-5.3 Instant ยท Line count comparison in responses on advice-style sample prompts.
29.2%
๐
5 May 2026๐ OpenAI (openai.com/index/gpt-5-5-instant/)
Result from OpenAI blog post example; full methodology not disclosed.