OpenAI's GPT-5.6 family Sol, Terra, and Luna went live on July 9, 2026, and became the preferred model powering Microsoft 365 Copilot the same day, across Word, Excel, PowerPoint, Copilot Chat, and Cowork. Each of the three tiers trades reasoning power for cost differently, so the model actually doing your work inside Copilot depends on which one gets selected.
Who this is for: Microsoft 365 users and IT decision makers trying to understand what changed in Copilot and which tier fits their work.
Who this is not for: readers looking for a deep technical breakdown of GPT 5.6's model architecture this is a practical usage guide, not a research paper summary.
What GPT 5.6 Sol, Terra, and Luna Actually Are
GPT-5.6 isn't one model it's three, built for different jobs rather than three quality levels of the same thing.
| Tier | Best For | Price (per million tokens, input/output) |
|---|---|---|
| Sol | Complex reasoning, long running agentic work | $5 / $30 |
| Terra | Everyday balanced work the default workhorse | $2.50 / $15 |
| Luna | Fast, high volume, simple tasks | $1 / $6 |
Early benchmark reporting put Sol roughly level with Anthropic's Fable 5 on raw coding intelligence, while using under half the output tokens and time, at a noticeably lower cost. That efficiency angle similar capability, meaningfully cheaper is the main story of this release rather than a raw intelligence jump.
How This Shows Up Inside Microsoft 365 Copilot
Microsoft and OpenAI worked together to optimize the GPT 5.6 family specifically for knowledge work across the Office apps, and the changes are already live.
- Word: turns rough ideas into more complete drafts with fewer rounds of back and forth prompting.
- Excel: runs deeper analysis and gets you from raw data to a usable answer faster.
- PowerPoint: builds a polished deck from a rough idea with less manual direction.
- Cowork: carries complex, multi step, cross-functional work with less hand holding between steps.
One detail worth knowing before you go looking for it: Copilot doesn't show you the names "Sol," "Terra," or "Luna" directly in most of the interface. Microsoft has been using its own generic labels, like "Quick Response" and "Think Deeper," instead so matching what you pick to the actual underlying model takes a bit of extra attention.
"Preferred" Doesn't Mean "Exclusive"
GPT 5.6 being the preferred model in Copilot doesn't mean it handles every single request. Microsoft has said availability and routing can vary by region and tenant configuration, and Copilot can select a model automatically based on what it judges the task needs. Where model selection is exposed, you can choose GPT 5.6 directly but "preferred" is a default, not a guarantee.
The family is also built to pull context from connected tools Microsoft 365, Slack, Notion, and Google Drive and hand back something closer to finished work rather than a rough first pass. That context pulling is part of what's driving the "fewer rounds of prompting" improvement across Word, Excel, and PowerPoint.
Which Tier Should You Actually Use
Sol makes sense when the task is genuinely hard: a long financial model, a complex multi document analysis, or agentic work that runs many steps unsupervised. Terra is the sensible daily default for most office work drafting, summarizing, first pass analysis. Luna fits high volume, low stakes tasks: quick rewrites, simple formatting, fast lookups where speed and cost matter more than depth.
Before rolling a tier out to a whole team, test it on your own work first. Take a handful of real tasks you do every week, run them through Sol, Terra, and Luna, and compare the output quality against the cost difference yourself rather than trusting a generic ranking.
Practical Takeaway
If your organization has Microsoft 365 Copilot, check which GPT 5.6 tier is actually active in your tenant rather than assuming it's the flagship Sol by default then run your own small side by side test across Sol, Terra, and Luna on real documents and spreadsheets before deciding which one your team should default to for everyday work.
No comments: