OpenAI rolled out GPT-5.6 to the public on July 9, 2026, ending months of speculation about the next generation of its language models. The lineup includes three distinct variants: Sol as the flagship, Terra for everyday work, and Luna as the most cost-efficient option.
GPT-5.6 Sol is the most capable of the three, built for tasks requiring deep reasoning across coding, science, and cybersecurity domains. Terra sits in the middle, designed as a balanced model for routine professional use. Luna rounds out the family with a focus on keeping token costs low while still delivering solid performance on common tasks.
GPT-Live: OpenAI’s New Voice Architecture
Alongside the text models, OpenAI also released GPT-Live, a full-duplex voice system that processes input continuously rather than waiting for turn-based exchanges. This means the model can listen, reason, and speak simultaneously, eliminating the awkward pauses that plagued earlier voice assistants.
In head-to-head testing, users preferred GPT-Live over the previous Advanced Voice Mode in roughly 75% of conversations. Expert science scores nearly doubled compared to the old system. The model can also delegate to a stronger reasoning model in the background when questions require deeper analysis, and it surfaces visual cards alongside spoken answers.
GPT-5.6 Now Powers Microsoft 365 Copilot
OpenAI confirmed that GPT-5.6 is now the preferred model within Microsoft 365 Copilot, meaning millions of enterprise users will interact with the new models through Word, Excel, PowerPoint, and Outlook without any manual upgrade. The shift happened quietly, with Microsoft updating the backend routing for Copilot users.
The release follows a period of tension between OpenAI and the US government. In early July, the company delayed the public rollout of GPT-5.6 after a request from federal officials, though the specific concerns behind that request were not publicly detailed. The eventual launch suggests those issues were resolved.
What Changed Under the Hood
Andrej Karpathy, former OpenAI researcher, delivered a talk at Microsoft Build that provided context on how models like GPT-5.6 actually work. His key point: LLMs do not try to succeed at tasks. They imitate patterns from training data. A model with roughly 80 transformer layers spends the same compute on each token as the human brain does on a single thought, but without any ambition or goal-directed behavior.
This distinction matters for anyone building on top of GPT-5.6. Understanding that the model pattern-matches rather than reasons means you can structure prompts and workflows more effectively.
Pricing and Availability
GPT-5.6 Sol is available through the OpenAI API and ChatGPT Plus subscriptions. Terra and Luna are accessible via the same channels, with Luna offering the lowest per-token cost for high-volume applications. Exact pricing tiers have not been publicly listed yet, but Anthropic’s Claude Opus 4.7 at $20/month and Cursor Pro at $20/month provide competitive reference points in the AI coding space.
FAQ
What are the differences between GPT-5.6 Sol, Terra, and Luna?
Sol is the flagship model optimized for complex reasoning tasks in coding, science, and cybersecurity. Terra is a balanced model for everyday professional work. Luna focuses on cost efficiency while maintaining acceptable performance for common tasks.
When was GPT-5.6 released to the public?
GPT-5.6 became generally available on July 9, 2026, following a limited preview period that started in late June.
Is GPT-5.6 available in Microsoft 365 Copilot?
Yes. OpenAI confirmed that GPT-5.6 is now the preferred model powering Microsoft 365 Copilot across Word, Excel, PowerPoint, and Outlook.
What is GPT-Live and how is it different from Advanced Voice Mode?
GPT-Live uses a full-duplex architecture that processes audio input continuously instead of turn-based exchanges. This allows the model to listen, think, and speak simultaneously, resulting in more natural conversations with fewer awkward pauses.
