On July 9, 2026, OpenAI closed out a short government-requested review window and pushed the GPT-5.6 launch across ChatGPT, the API, and Codex within 24 hours. The rollout hit every paid tier at once — Plus, Pro, Business, and Enterprise accounts got access first, with Free and Go tiers following inside the same window. For a lineup that used to ship as one model, GPT-5.6 arrived split into three sized versions, Sol, Terra, and Luna, each built for a different job rather than just a different price bracket.
In this article
- What changed across the three new GPT-5.6 model tiers
- How the rollout reached ChatGPT Plus, Pro, and Free accounts differently
- Why Microsoft named GPT-5.6 its preferred engine for 365 Copilot
- What the new ChatGPT Work agent does with GPT-5.6 underneath it
- What this actually means for a one-person ChatGPT habit
OpenAI’s GPT-5.6 Launch Splits One Model Into Three Sized Versions
The headline change in the GPT-5.6 launch is structural: OpenAI replaced its previous single flagship with three separate models built for different jobs, not just different price points. Sol is the largest and most capable, positioned as OpenAI’s strongest model yet for long-horizon security work such as vulnerability research, and reported as meaningfully more token-efficient on agentic coding tasks than its predecessor. Terra is the everyday-work version, tuned to balance quality against cost. Luna is the fastest of the three, built for high-volume, low-latency requests rather than depth.
That three-way split matters more than any single benchmark number. Instead of one model trying to be fast, cheap, and capable at once, OpenAI now routes each request to a different tier by default, depending on which plan and which product sent it.
The GPT-5.6 Launch Reaches ChatGPT Plus Before It Reaches Free Users
Not every ChatGPT account received the same version on day one. Plus, Pro, Business, and Enterprise users can run GPT-5.6 Sol at medium-effort settings and above, and Pro and Enterprise accounts get an additional Pro-effort mode reserved for the hardest tasks — the same tier math shows up whenever ChatGPT’s $200 plan gets stacked against Claude’s top-end plan. Free and Go accounts, along with ChatGPT Work and Codex by default, start on Terra instead.
The access split looks like this:
- Plus, Pro, Business, Enterprise: GPT-5.6 Sol at medium effort and above
- Pro and Enterprise only: a higher Pro-effort mode for the hardest requests
- Free, Go, and Work/Codex defaults: GPT-5.6 Terra
- All paid tiers: a manual switch between Sol, Terra, and Luna, each with its own effort dial
Microsoft Made GPT-5.6 the Default Engine Behind Word and Excel
Microsoft 365 Copilot switched its default engine to GPT-5.6 the same week OpenAI shipped it, covering Word, Excel, PowerPoint, Chat, and Cowork. Microsoft’s announcement frames the swap around fewer prompting rounds to reach a usable draft in Word and more efficient token use for deeper analysis in Excel. Microsoft still runs Anthropic, Meta, and Mistral models elsewhere in its data centers, so “preferred” describes a default, not an exclusive arrangement.
A model swap that changes your default answer without changing your subscription price is the quiet kind of update — no new tier to buy, just a different engine running under the same plan.
A New ChatGPT Work Agent Rides Along With the GPT-5.6 Launch
OpenAI didn’t only ship a smarter model with the GPT-5.6 launch; it shipped a new agent built to run on top of it. ChatGPT Work is built to:
- Gather context across connected apps and files
- Produce finished documents, spreadsheets, presentations, and reports
- Keep working on a single task for hours instead of answering in one turn
The rollout starts with Pro, Enterprise, and Edu accounts on the Mac and Windows apps, then extends to Plus and Business over the following days, with web access to follow.
For a Solo Consultant, the Real Change Is Effort Settings, Not Hype
For a one-person practice, the GPT-5.6 launch changes less about what ChatGPT can do than about which dial is set when you use it. My own ChatGPT habit stays modest — quick questions, email headlines, fast reframes on a weekly basis — the kind of task Terra or Luna likely already handles by default, the same pattern that showed up the last time OpenAI swapped ChatGPT’s default model. If I routed a slower deliverable through ChatGPT Work instead of drafting it by hand, the product description suggests it would keep working after I closed the laptop, which is a claim worth testing on a low-stakes file before trusting it on a paying one.
For me, the GPT-5.6 launch is mostly a reminder to check which model answered before judging the answer. A quick reframe or a subject line probably runs on Terra or Luna without me ever touching a setting, and that covers most of what I send through ChatGPT in an average week. ChatGPT Work stays on the list to watch rather than recommend, since letting an agent run unattended on client files is a different risk than a five-second question. The two minutes it takes to check the model name under a reply is worth spending before the next deadline.
Sources
- GPT-5.6: Frontier intelligence that scales with your ambition — OpenAI
- GPT-5.6 is now the preferred model in Microsoft 365 Copilot — OpenAI
- OpenAI launches its new family of models with GPT-5.6 — TechCrunch
- OpenAI releases GPT-5.6 and ChatGPT Work tool — Axios
- OpenAI to publicly release GPT-5.6, rolls out conversational AI models — CNBC
AI-assisted research and drafting. Reviewed and published by ToolMint.