OpenAI began rolling out Dreaming V3 on June 4, 2026, a background process that automatically synthesizes ChatGPT's memory from conversations without requiring explicit instructions. The system represents OpenAI's most significant memory upgrade since the original ChatGPT launch, replacing the saved-memories list with an automatic, temporally-aware foundation.
Automatic Memory Synthesis With Temporal Awareness
Dreaming V3 synthesizes conversation context across three dimensions: freshness, continuity, and relevance. The system automatically updates memories as time passes—for example, revising "You're going to Singapore in July" to "You went to Singapore in July 2026" after the trip concludes. This temporal awareness marks a significant improvement over static memory systems that require manual updates.
Performance Improvements and Efficiency Gains
OpenAI reports factual recall rising from 41.5% in 2024 to 82.8% in 2026 on internal evaluations, with preference and time-sensitive scores reaching the low-to-mid 70% range. A 5x reduction in compute requirements makes the system practical for Free tier users, who gain access to Dreaming for the first time. The efficiency gains suggest meaningful architectural improvements in the underlying memory synthesis process.
Rollout Schedule and User Controls
Dreaming V3 began rolling out to ChatGPT Plus and Pro users in the United States on June 4, 2026. OpenAI plans to expand to additional countries and Free and Go users over the coming weeks. The system includes a transparency surface that shows what ChatGPT knows and allows users to correct, dismiss, or instruct the system—addressing privacy and control concerns.
Shift to Ambient Intelligence
The automatic synthesis represents a shift from explicit memory management to ambient intelligence that tracks user context over time. Users no longer need to ask ChatGPT to remember specific details; the system extracts and updates relevant information from ongoing conversations. This approach reduces friction while maintaining user control through the transparency interface.
Key Takeaways
- OpenAI released Dreaming V3 on June 4, 2026, automatically synthesizing ChatGPT memory across conversations
- Factual recall improved from 41.5% in 2024 to 82.8% in 2026 on internal evaluations
- The system features temporal awareness, automatically updating memories as events pass
- A 5x compute reduction enables rollout to Free tier users for the first time
- Users can view, correct, and control memories through a new transparency interface