Amp retires the Dial and replaces it with a model selector

Amp has retired the Dial. The four modes that sat on it — low, medium, high and ultra — are gone. In their place is a model selector. You choose the model a thread runs on and the effort it works at, you can change both in the middle of a thread, and the selector shows whether a subscription, an API key or Amp credits will be billed.

Amp set that out in a post titled “Bye Bye, Dial.” Nicolay Gerold, a member of technical staff at Amp, posted the same change on X.

Bye bye, Dial.

low, medium, high, and ultra are retired. In their place is a model selector: pick your model, switch mid-thread, and see exactly who gets billed.

https://t.co/l8HrxVEC0P https://t.co/lCYdUgMvOx

— Nicolay Gerold (@nicolaygerold) November 12, 2026

For most of the year Amp had argued it should not have a model selector at all. In January it published a page under the title “Amp Does Not Have a Model Selector,” and said an agent works better as a product built end to end than as a kit people assemble themselves. The job, it wrote, was to build the harness — the system prompt, the tools, the environment, the subagents — around a specific model, and it could do that job better if Amp was not used in a thousand mismatched ways. In July it shipped the Dial so you could think about how hard a task was, rather than about model names and version numbers. Amp says that part worked: when it changed the default model behind the Dial, nobody complained.

What broke that design was billing. Once people could bring their own subscriptions and API keys, and Amp became free when they did, the selected model started to decide who paid. Amp’s post says a thread on medium could use your Claude plan, your ChatGPT plan or Amp credits, and it became very hard to tell which. The Dial also picked up a pile of settings to cover every combination. So the company says it has stopped hiding the model.

You are still not picking a bare model. Amp says it builds a harness around each one: the system prompt and the tools are optimised for that model, and both switch when you switch models in a thread.

Switching mid-thread, and what it costs

On the Dial, a thread kept the mode, and the model, it started with. Amp did that because, in its experience, a model became confused when it continued a conversation another model had started, and the results were never good. It now says models are good enough to recover from that.

You open the selector, make the change, and the new model answers your next message. The transcript records it, for example “Switched to GPT-6.1 Sol · High.” In the web app the selector is the model button below the prompt box, or Ctrl+S. The CLI uses the same shortcut. In Amp’s own product screenshot, Staff Pick is listed first as Opus 5.5, and the card beside it describes Claude Opus 5.5 as served by a personal Claude Subscription.

A switch still has a price. Prompt caches belong to one model, so the new model reads the whole thread uncached, and the first message after a switch costs more. A thread that is running an agent from a plugin, or Claude Code, keeps that agent. You cannot switch that thread to another model. Only the thread’s owner can switch.

Quinn Slack, Amp’s CEO, called the change a “huge simplification,” and singled out mid-thread switching. “I’ve never wanted this, but I also don’t like mustard or cilantro and yet millions do.”

Huge simplification in Amp shipped today.

And now you can change models mid-thread. I’ve never wanted this, but I also don’t like mustard or cilantro and yet millions do. https://t.co/2vlsNK6Od0

— Quinn Slack (@sqs) November 12, 2026

Staff Pick, and what happens to your Dial

New threads start on Staff Pick by default, which is the model the Amp team would choose for most work. Amp names that model as Claude Opus 5.5, and says it will change the pick when another model does better. The exception is if you had already set a default mode on the Dial. In Settings → Model Selector, under New Threads Start On, you can also start new threads on the model you picked last, or on one fixed model. A thread that has already started keeps its model when the Staff Pick changes.

Existing threads keep working. A thread you started in high continues on the model it was already running. Dial settings do not carry over. If you changed the model behind a mode in Tune Modes, you pick that model in the selector now. If you put your own agents on the Dial in Build Dial, Amp’s post says you will find them under Other Models. The docs put those agents in a Your Agents group at the top of that list.

If you never changed the Dial’s settings, these are the models Amp says your modes ran. The first three now sit at the top of the selector, and Claude Fable 5.2 is under Other Models:

Former Dial mode	Model it ran
low	GLM-5.3-Flash
medium	Claude Opus 5.5
high	GPT-6 Astra
ultra	Claude Fable 5.2

You can pin models in Settings → Model Selector, or leave the selector on Staff Pick. The docs also describe Top Models, a list of up to eight models below Staff Pick; an Effort control; and a Speed row that switches between Standard and Fast, and Plaid for the models that offer it. Each row of the selector carries an icon for the connection that will be billed. Model Routing is still what decides which of your connections serves a model. A workspace admin can set New Threads Start On, Default Speed and Top Models for members, and enterprise admins can turn models off in Allowed Models.

In the CLI and the SDKs, --model (or the model option) picks a model for a new thread by its ID, written provider/name, instead of --mode. --effort sets the effort. Amp’s docs give this example:

amp -x "Review this repository" --model openai/gpt-6-astra --effort xhigh

--mode also still accepts a model ID. Beyang, who works on Amp and previously founded Sourcegraph, wrote that there are “so many good models to choose from now and increasingly token cost is important as users route between different subscriptions.” Thorsten Ball, who also works on Amp, put the reversal more dryly: “Probably the biggest news in AI today: Amp now has a model picker.”

Probably the biggest news in AI today:

Amp now has a model picker. https://t.co/SlCzkYtEwC

— Thorsten Ball (@thorstenball) November 12, 2026