OpenAI Releases GPT-5.5 Instant with Enhanced Accuracy

OpenAI Deploys More Accurate GPT-5.5 Instant Model

OpenAI announced the release of GPT-5.5 Instant, a new default model for ChatGPT designed to provide greater accuracy and more concise responses. The update aims to significantly reduce factual errors and hallucinations in the AI's output.

Key Performance Metrics and Upgrades

Internal evaluations show the new model produces 52.5% fewer hallucinated claims on high-stakes topics like medicine and finance. Performance on academic benchmarks also saw notable gains, with math competition accuracy rising to 81.2% from 65.4% in the prior version. Additionally, responses are now more direct, using over 30% fewer words on average while maintaining conversational quality.

New Personalization Features and User Access

The update introduces "memory sources," allowing users to see and control the data informing personalized responses. The model can now also search past conversations and connected Gmail accounts for tailored answers. GPT-5.5 Instant is rolling out to all users as the new default, with paid subscribers retaining access to the previous version for three months before it is retired.

Summary

The launch of GPT-5.5 Instant marks a significant step toward more reliable and efficient AI interactions. Users can expect improved factual consistency and a more streamlined conversational experience, with enhanced personalization features rolling out progressively across different user tiers.

FAQ

Q: What is the main improvement in GPT-5.5 Instant?
A: The primary improvement is enhanced accuracy, with internal tests showing a 52.5% reduction in hallucinated claims on critical topics and a 37.3% drop in inaccurate claims overall.

Q: Is the new model available to all ChatGPT users?
A: Yes, GPT-5.5 Instant is now the default model for all free and paid ChatGPT users, and it is also available in the API.

Source: Investing.com