OpenAI released a series of major updates yesterday around its GPT-5.6 Sol model, emphasizing efficiency gains and broader access to advanced tools for researchers.

ARC-AGI-3 Improvements

The company detailed how simple API changes, including retained reasoning and context compaction, tripled scores on the ARC-AGI-3 benchmark while using 6x fewer output tokens. The model moved from struggling with open math problems to stronger performance.

Self-Optimization Efficiency

OpenAI reported 20% lower serving costs from GPU kernel improvements and 15%+ better token-generation efficiency from improved speculative decoding. These optimizations contribute to a better cost-intelligence curve.

Free Access for Researchers

Additionally, the company announced the "ChatGPT for Academic Researchers" program, granting free access to frontier models for an initial 10,000 researchers, expanding to 100,000 by 2027. The access includes business-grade privacy and does not use data for training by default.

Greg Brockman, President and Co-Founder of OpenAI, noted that “Science moves faster when more researchers have access to the best tools.”

The conversation on X centered heavily on these developments, with millions of views on the official posts.

The Bottom Line

OpenAI continues to focus on efficiency improvements and democratizing access to advanced technology, a move expected to impact the research and development community in the near term.