Dume Now Runs on GPT 5.6: Meet Luna, Terra, and Sol

Dume acts on your inbox, your calendar, and your files, not just your questions. That only works if the model behind it can tell the difference between a two-second lookup and a task that needs real thinking. Dume's AI core just moved to a new model family built around exactly that distinction: GPT 5.6 Luna, GPT 5.6 Terra, and GPT 5.6 Sol, replacing a single general-purpose model with three tiers tuned for different points on the speed-versus-depth trade-off.

• Dume now runs three GPT 5.6 tiers, Luna, Terra, and Sol, instead of one general model
• Luna handles fast, lightweight requests; Terra covers everyday assistant work; Sol takes on the heaviest tasks, like multi-site research in Dume Cowork
• GPT-5.1 Codex Mini is retired. GPT 5.6 Luna is its direct successor, and any workflow pinned to Codex Mini now runs on Luna automatically
• Model access still follows your plan, the same way it always has on Free, GO, PRO, and MAX

Why Move to Three Tiers Instead of One?

A model doing your morning briefing and a model comparing prices across three shopping sites in Dume Cowork are not doing comparable work. One is a quick summary; the other is a multi-step browsing session that can run for minutes. Forcing both through the same model means either the quick task waits too long, or the hard task doesn't get enough reasoning to finish it right.

This isn't a new idea for Dume specifically, it's an extension of how the product already works. Free, GO, PRO, and MAX already separate “Basic,” “Standard,” “Premium,” and “All Frontier” AI models by plan. Splitting the model layer itself into three named tiers, Luna, Terra, and Sol, just makes that existing trade-off visible and controllable instead of buried in a single label.

What Is GPT 5.6 Luna Built For?

Luna is the tier built for speed, and it's the one running quietly behind requests where waiting on deep reasoning would just slow you down: a quick WhatsApp reply, an inline rewrite through Dume for Chrome, a short lookup in chat. If the request is well-defined and doesn't need multi-step planning, Luna is almost certainly the model doing the work.

Luna is also the direct successor to GPT-5.1 Codex Mini, which has now been fully retired. If you had lightweight coding or quick-turnaround tasks pinned to Codex Mini, they route to Luna automatically. Nothing to reconfigure on your end.

What Is GPT 5.6 Terra Built For?

Terra sits in the middle, and it's where most day-to-day assistant work will land, things like Email Triage sorting your inbox, a Morning Briefing pulling together your meetings and messages, or drafting a reply that needs to sound like you wrote it. Terra trades a little of Luna's speed for meaningfully stronger reasoning, which matters once a task has more than one moving part.

If you're not sure which tier fits, Terra is the safer default. It's built for exactly the kind of mixed, moderately complex work that makes up most of what an AI executive assistant actually does in a day, not the simplest requests, not the hardest ones.

What Is GPT 5.6 Sol Built For?

Sol is the flagship, reserved for the requests where depth matters more than speed: a Dume Cowork session comparing options across several sites before booking something, pulling a report together from Notion and Google Docs sources, or working through a task with real consequences if it gets the details wrong. Sol costs more to run than Luna or Terra, and that's the direct trade-off for the extra reasoning it brings.

Reach for Sol when a task is genuinely hard, not by default, the same way you wouldn't hand a two-line email to your most senior analyst.

Which Plan Gets Which Model?

Model access on Dume has always followed your plan, not the other way around. Free and GO plans include what the pricing page calls “Basic” and “Standard” AI models; PRO steps up to “Premium” models; MAX unlocks “All Frontier” models, the newest and most capable tier available. GPT 5.6 Luna, Terra, and Sol extend that same tiered structure, so a MAX plan built around unlimited agents, 10 MB of memory, and every MCP connector also gets access to the deepest reasoning tier.

That gating isn't new with this release, it's the same shape the plan lineup has used since GO, PRO, and MAX launched, with model access scaling alongside credits, agents, and memory the higher you go. Check your plan's exact model access on the pricing page for the specifics that apply to your account.

What Happened to GPT-5.1 Codex Mini?

GPT-5.1 Codex Mini has been deprecated and is no longer available. Its replacement is GPT 5.6 Luna, which takes over the same fast, lightweight role with an updated model underneath it. Existing conversations, routines, and workflows carry over automatically. There's genuinely nothing you need to touch.

How to Choose the Right Model

A simple way to think about it:

• Pick Luna for quick, low-stakes requests, chat replies, inline edits, short lookups.
• Pick Terra for most everyday work: Email Triage, Morning Briefing, drafting, and routine automation. It's the default for a reason.
• Pick Sol for the hard cases: multi-site Dume Cowork sessions, long reports, anything where you'd rather wait a bit longer for a better answer.

Where Dume supports manual model selection, you can set it directly. Otherwise, the assistant chooses based on the request, the same way it already decides which integration or tool to reach for.

Frequently Asked Questions

Try It Now

The GPT 5.6 lineup is live today across Dume's web app, WhatsApp, and Dume Cowork. No update or setup required, the new models are already powering your assistant on whichever plan you're on.