OpenAI announced today that its latest model, GPT-5.5 Instant, is now available to paying users worldwide, with a full rollout to free users scheduled for tomorrow, June 26. This significant upgrade marks the largest platform enhancement since 2026 and will replace the previous GPT-5.3 Instant as the default model across all ChatGPT platforms, supporting the daily conversations of hundreds of millions of active users.
In terms of performance, OpenAI reports notable improvements. The new model shows a 52.5% reduction in hallucinations—erroneous or fabricated information—in high-risk areas such as healthcare, legal, and financial sectors. Additionally, instances of users flagging responses as factually incorrect have decreased by 37.3%. Standardized testing results indicate substantial gains: accuracy on the AIME 2025 math competition jumped from 65.4% to 81.2%, while GPT-4.Science questions achieved an increase from 78.5% to 85.6%. Multimodal reasoning tasks also saw improvement, with MMMU-Pro scores rising from 69.2% to 76.0%. Many of these metrics now closely resemble the levels of flagship models from two years ago.
User experience has also been a key focus. The GPT-5.5 Instant offers more direct and concise responses by reducing unnecessary courtesies and overly segmented outputs. Compared to previous versions, the model delivers around 30% fewer words and nearly 29% fewer lines, addressing past criticisms of verbose replies and redundant formatting. It also better handles image comprehension, STEM question answering, and internet search judgment, leading to higher accuracy in everyday queries, technical writing, translations, and operational guidance.
One noteworthy feature of the new model is its built-in intelligent routing system. This mechanism automatically assesses the complexity of user requests, ensuring swift responses for simple questions while seamlessly shifting to deeper reasoning modes for more complex tasks—users won’t need to manually switch modes to receive appropriate processing.
Regarding context window capacity, AI models now follow a layered tiered approach: free users get 16,000 tokens, ChatGPT Plus and Business users get 32,000 tokens, and Pro and Enterprise members can access up to 128,000 tokens. This differentiated setup aims to balance accessibility with enhanced functionality for paid tiers, providing a clear incentive for users to upgrade.