Models

Pick a model for a conversation, let Neoo route automatically, keep favorites, add custom models and set reasoning effort.

Each conversation runs on one model. You choose it in the message bar, and you can change it between turns.

Open the model picker

Click the model name below the message field. The picker has three parts: a column of sources on the left, the model list in the middle and, when the conversation lets you set it, the Effort column on the right. Speed and Ultracode sit at the bottom.

The sources on the left are:

  • Favorites (the star): your starred models. It appears once you star a model.
  • One logo per provider. A greyed-out logo means the provider isn't ready yet.
  • Automatic: lets Neoo pick the model for you.
  • All models / manage: the gear icon at the bottom, to organize the list.

A short bar marks the source you're looking at. Each model row shows its provider's logo, a check mark on the model in use and a New badge on recent models.

Type in Search models to filter the current source, then press Enter to pick the first result. With the keyboard, the left arrow moves from the search field to the sources, the right arrow to the Effort column, and the up and down arrows move inside a column. Tab moves from one source to the next.

If a provider isn't connected, its tab explains why and shows a button such as Sign in or Install that opens a terminal with the right command. See Connect your AI accounts.

Keep favorites

Hover a model and click the star to Add to favorites. Favorites from every provider are gathered in the Favorites tab. Click the star again to Remove from favorites.

Find free models

The OpenRouter and OpenCode tabs have filters:

  • All: every model.
  • Tools: models that can use tools.
  • Free: free models only.

GPT-6.2 Astra and GPT-6.2 Sol

The Codex tab lists GPT-6.2 Astra and GPT-6.2 Sol, released by OpenAI on November 17, 2026. Both offer Low to Ultra reasoning effort and the Fast speed. They are available to ChatGPT Plus, Pro, Enterprise and Edu accounts; if your account doesn't have them yet, OpenAI rejects the turn and Automatic account selection tries another connected Codex account.

OpenAI cyber models (Daybreak)

OpenAI's Daybreak models for security work (Daybreak Blue and Daybreak Red) are reserved for approved ChatGPT accounts. They appear in the Codex tab, marked cyber, only when at least one of your connected Codex accounts receives them from OpenAI. Approval alone isn't enough: Daybreak is turned off by default, so turn it on for your account first, then run one Codex turn so Neoo can read the account's updated model list.

When you pick a Daybreak model, Automatic account selection runs the turn on an account that has it. If you pinned an account without access, OpenAI rejects the request. These models offer Low to X-High reasoning effort.

The Daybreak models appear only in the desktop app for now. Remote, web and mobile don't list them yet.

Add a custom model

  1. Open a provider tab and click All models / manage.
  2. Type the model ID into Enter a model ID.
  3. Click Add custom model.

In the same mode, you can move models up or down, choose Hide from picker, and use Remove custom model.

Reasoning effort and speed

When the selected model supports it, the Effort column on the right lists its reasoning levels, for example Automatic, Low, Medium, High or Max. Click a level to use it: the selected level has a check mark, and Default marks the model's own default. Higher levels think longer on hard problems. Claude Sonnet 5.5 and Claude Haiku 5.5 offer Low to Max but not Off: Claude Code can't reliably turn their thinking off, so use Low for the fastest, cheapest replies.

When you pick another model that has effort levels, the picker stays open so you can set its effort. Click outside or press Escape to close it.

Some models come in several context window sizes. Choose one under Context, below the effort levels.

The Speed control at the bottom of the picker has up to three choices:

  • Standard: the provider's normal tier.
  • Fast (one lightning bolt): the provider's fast tier, when the model offers it.
  • Ultrafast (two lightning bolts): appears for a Codex model when your ChatGPT account actually receives OpenAI's Ultrafast tier for that model (today, GPT-6.2 Sol on eligible plans). Neoo reads this from your account's own Codex model list, so when OpenAI opens it to more models, the option follows on desktop, Remote, mobile and web without waiting for a Neoo update. With Automatic account selection, the turn runs on a connected Codex account that has Ultrafast. Ultrafast uses much more of your limits and credits; if OpenAI refuses it, the turn stops with an error and you can switch back to Standard or Fast. You can't switch to or from Ultrafast while a turn is running.

For Claude models that support it, Ultracode sits next to Speed.

The message bar shows the current effort next to the model name, with one lightning bolt for Fast, two for Ultrafast, and an icon when Ultracode is on.

Neoo remembers the effort and speed you last used with each model. When you come back to a model, in the same conversation or a new one, its settings come back with it.

Switch provider or model during a conversation

You can change the model, or move to another provider, between two turns. If you change it while the agent is still working, the current turn finishes on the old model and the change applies to your next message.

  • Same provider, other model: the conversation continues in the same session. A Model changed card shows the old and new model.
  • Other provider or other account: sessions can't move between providers, so Neoo starts a new session and gives it the conversation so far. A Continued with … card shows how many messages were carried over.

Short conversations are carried over intact. Long ones keep your original request and the most recent messages. The card then says how many older messages were left out, and the agent can read them on request. Each reply keeps the name of the provider and model that wrote it. The agent's private reasoning, tool output and older attachments are never carried over.

If a provider can no longer reopen its saved session, Neoo starts a fresh one with the saved conversation and shows Native session unavailable.

To choose how much history is carried over, go to Settings → General and set History carried over on a switch to Compact, Standard or Extended.

Automatic routing

With Automatic, routing rules pick the model for each turn. On the first turn of a new conversation, they also choose the provider, which then stays the same. Choosing a specific model pins the conversation to it.

On that first turn, interface and design work prefers Claude, and backend or technical work prefers Codex. This is only a preference: if the preferred provider isn't connected, or none of its models is allowed, Automatic picks among your other connected providers instead of stopping.

The Automatic tab appears when automatic routing is turned on. To set it up, go to Settings → Plugins, open Automatic routing and click Manage.

  • Preset: Off, Economical, Balanced or Quality. Economical favors the cheapest models for simple work. Quality never downgrades.
  • Rules: click Add a rule, give it an Id, describe when it applies under Criterion, then pick a Provider and Model. Optional Paths limit the rule to requests that name matching files.
  • Project file: click Create the rules file to keep rules in the project, versioned with its repository.
  • Allowed providers and models: limits what Automatic may choose on this computer, including requests from your phone and the web. Click Save selection when you are done. Models you pick by hand stay available.

Rules always target the newest version of their model family. A rule written for claude-opus-5 runs on Opus 5.5, and one written for gpt-5.6-sol runs on GPT-6.2 Sol. Neoo never switches tier (Opus stays Opus, Sol stays Sol) and never downgrades. A model newly announced by your provider is picked up as soon as Neoo reads its model list. If Allowed providers and models is restricted, Automatic uses the newest version you allowed, so add a new model there to let routing use it.

Routing never switches provider in the middle of a conversation. When no rule applies, your current choice stays in place.

Team workspaces

In a Team workspace, the models and accounts you can use depend on the team's configuration. Check the model shown in the message bar before you send.

Next steps