OpenAI broadly released GPT-5.6 in three variants, Sol, Luna, and Terra, alongside ChatGPT Work and a new default model inside Microsoft 365 Copilot.
By Quantara Editorial Team · AI Tools · Published · 7 min read
OpenAI broadly released GPT-5.6 on Thursday, July 9, 2026, after staggering the model's rollout at a US government request. Rather than a single model, OpenAI shipped a three-model family built around a shared architecture but tuned for different jobs: Sol, Luna, and Terra.
What Are Sol, Luna, and Terra?
Sol is the most powerful variant, aimed at demanding agentic and coding workloads. Luna is built for speed, trading some depth for lower latency on high-volume tasks. Terra sits in between, positioned as the everyday default for general work. CEO Sam Altman told CNBC that Sol is 54% more token-efficient on agentic coding tasks than its predecessor, framing the improvement around cost: "every enterprise now is thinking about spend and the value they're getting in exchange for AI," he said.
ChatGPT Work Launches Alongside GPT-5.6
OpenAI paired the model release with ChatGPT Work, a product built on GPT-5.6 that pulls context from a team's existing tools to turn scattered notes, drafts, and half-finished ideas into completed work, while keeping projects moving and the user in control of what ships.
GPT-5.6 Becomes the Default in Microsoft 365 Copilot
Microsoft announced that GPT-5.6 is now the preferred model inside Microsoft 365 Copilot, spanning Word, Excel, PowerPoint, Chat, and Cowork. That makes GPT-5.6 the default reasoning engine behind a large share of enterprise Copilot usage rather than an optional add-on.
Smaller ChatGPT Changes Worth Knowing
- Custom-instruction storage increased from 1,500 to 5,000 characters for Plus, Pro, Enterprise, Business, and Education users, giving more room to define tone and behavior once instead of repeating it per chat.
- A new unified search now covers chats, projects, images, and documents from the ChatGPT sidebar on web, iOS, and Android, with filters to narrow results by content type.
What This Means in Practice
The three-tier split is OpenAI's clearest acknowledgment yet that "one model for everything" is giving way to task-matched deployment: Sol for work where reasoning depth pays for itself, Luna where latency and cost dominate, Terra as the safe default. Teams building on the API should benchmark their own workloads across the three rather than assuming Sol is automatically the right choice — the efficiency gains are concentrated in agentic coding, and may not translate evenly to every use case.
Build Prompts That Work Across the Family
Use Verbito's AI Prompt Generator to structure prompts around the task, not the model, so the same instructions carry over cleanly whether you route a job to Sol, Luna, or Terra.
Sources
- TechCrunch: OpenAI launches its new family of models with GPT-5.6
- Axios: OpenAI releases GPT-5.6 and ChatGPT Work tool
- Nextgov/FCW: OpenAI's advanced GPT-5.6 models to be publicly released
- OpenAI: Newsroom
Frequently asked questions
- What is the difference between Sol, Luna, and Terra in GPT-5.6?
- Sol is the most powerful variant, tuned for demanding agentic and coding work. Luna is optimized for speed and lower latency. Terra is the balanced, everyday default. All three ship as part of the same GPT-5.6 release rather than as separate model generations.
- Is GPT-5.6 available to everyone now?
- OpenAI broadly released GPT-5.6 on July 9, 2026, after an initial staggered rollout requested by the US government. It is available through ChatGPT and the API, and it is now the preferred model inside Microsoft 365 Copilot.
- What is ChatGPT Work?
- ChatGPT Work is a product powered by GPT-5.6 that gathers context from a team's tools to turn scattered notes and drafts into finished output, while keeping the user in control of what gets shipped.
- How much more efficient is Sol than the previous model?
- OpenAI CEO Sam Altman told CNBC that Sol is 54% more token-efficient on agentic coding tasks compared with its predecessor. That figure is specific to agentic coding workloads and should be validated against your own use case before assuming it generalizes.