Appearance
Models & reasoning cheat sheet
Every agent runs on a model you pick from MyChatBot's model catalog. This page covers what's in the catalog, how reasoning depth works, how to switch an agent's model in the app, and the gotchas that bite when you do.
Cheat sheet
| Concept | What it is | Where you set it |
|---|---|---|
| Model catalog | Dozens of models across many providers — Gemini, Claude, GPT, Grok, DeepSeek, Qwen, Llama, and more | The Model dropdown on an agent |
| Default model | The model an agent uses until you change it. Today that's Gemini 3 Flash — cheapest and fastest | Shown with a · default tag in the dropdown |
| Per-agent model | Each agent runs on the model chosen in its Settings → Identity → Model dropdown | Agent → Settings tab → Identity → Model |
| Reasoning depth | How much a model "thinks" before answering. Higher-reasoning models come as their own entries in the picker | Pick a model (see below) |
| Think step by step | A toggle that adds explicit think/analyze steps before the answer — most useful on cheaper models | Agent → Settings tab → Tools |
In the app
Open your agent from https://app.mychatbot.app/agents, go to the Settings tab, and find the Model dropdown in the Identity card. Whatever you pick there is the model that agent runs on — for chats and for scheduled tasks alike.
Reasoning depth
Some models "reason" more before they answer. In MyChatBot, reasoning depth is a property of the model you pick, not a separate slider. Higher-reasoning variants appear as their own entries in the picker — for example, a plain Gemini 3.5 Flash and a Gemini 3.5 Flash (High) that thinks harder for a bit more cost.
One family, several entries
A model family can show up as more than one option — a standard entry and a higher-reasoning entry. If you want deeper reasoning, choose the higher-reasoning entry rather than looking for a hidden dial; the depth is baked into the option you select.
Separately, there's a Think step by step toggle (under Settings → Tools). It scaffolds explicit "think" and "analyze" steps before the final answer — a nudge that helps smaller, cheaper models decompose a hard question. Frontier reasoning models (Claude Opus, Claude Sonnet with thinking, Gemini Pro) already do this internally, so turning it on there is mostly overhead.
Representative models
The picker always shows the live catalog — the exact list can change as models are added or retired. A few common picks:
| Model | Pick it when |
|---|---|
| Gemini 3 Flash | Default. Cheapest + fastest; good for most work. |
| Gemini 3.1 Pro | Long-context reasoning and multi-step plans. |
| Claude Opus 4.6 | Premium reasoning — nuanced writing, code review, sustained planning. |
| Claude Sonnet 4.6 | Balanced; drafting copy when Flash is too light. |
| GPT-5 | Tool-heavy sessions and code. |
| GPT-5 mini | High volume — classification, routing, short summaries. |
| GLM-5.2 | You want maximum built-in reasoning effort. |
| Gemini 3.5 Flash (High) | A high-reasoning Flash variant without paying for a heavier tier. |
Picking a model for your agent
Every agent starts on the default model (Gemini 3 Flash) until you change it. To switch:
- Open the agent from https://app.mychatbot.app/agents.
- Go to the Settings tab.
- In the Identity card, open the Model dropdown.
- Choose a model — the option list shows each model's display name (the default is tagged
· default), and a short description appears under the dropdown once selected. - Save. The agent's next run uses the new model.
Any model works with any agent
Every entry in the picker is ready to use on any agent — there are no unsupported combinations to worry about.
Team directors don't have a model of their own
A team director routes work to its members, so its Settings tab shows the team roster instead of a Model dropdown. Set the model on each member agent instead.
Gotcha: switching models changes reasoning depth
Because reasoning depth is part of the model, switching to a lighter model lowers how hard the agent reasons — even if the wording of your instructions is identical.
A model switch resets reasoning to the new model's level
Switch an agent from a high-reasoning model (e.g. GLM-5.2 or Gemini 3.5 Flash (High)) to a standard one (e.g. Gemini 3.1 Pro) and the next answers will reason less. To keep deep reasoning after a switch, pick another model that also reasons deeply — check the target model's description before you switch, not after.
Retired models are repointed for you
Occasionally a model is retired by its provider. When that happens MyChatBot maps it to a successor automatically, so an agent you configured earlier keeps working without you re-picking anything.
Scheduled tasks use the agent's model
A scheduled task has no model setting of its own — it runs on the agent's model, the one in Settings → Identity → Model, re-read each time the task fires. To change what a scheduled task runs on, change the agent's model there. See Tasks & schedules.
Token spikes right after toggling reasoning
Turning reasoning on or off mid-conversation can make the next reply heavier on tokens. MyChatBot automatically compresses a long conversation and retries if a turn runs over, so it recovers instead of failing — but expect one bigger run right after the change.
Default model per agent type
Different agent types ship with a sensible default so a new agent is useful out of the box. You can change any of them from the Model dropdown.
| Agent type | Default model | Change it? |
|---|---|---|
| Platform assistant | GPT-5 | Yes — Settings → Identity → Model |
| Personal assistant | Claude Opus 4.6 | Yes — Settings → Identity → Model |
| Skill architect | Claude Opus 4.6 | Yes — Settings → Identity → Model |
| Agents you build yourself | Claude Sonnet 4.6 | Yes — Settings → Identity → Model |
The default is just a starting point you can move — leaving the field on the default keeps the platform's recommended model, and picking anything else overrides it for that agent only.
Best practices
Do
- Start on the default (Gemini 3 Flash) and step up only when answers feel shallow.
- Reach for a higher-reasoning entry (e.g. Gemini 3.5 Flash (High), GLM-5.2) when you need deeper thinking without jumping to a premium tier.
- Turn on Think step by step on cheaper models that need a nudge to break problems apart.
- Re-check the target model's description whenever you switch an agent's model.
Don't
- Don't assume reasoning depth survives a model switch — it doesn't; it follows the model you land on.
- Don't stack Think step by step on top of a frontier reasoning model (Opus, Sonnet with thinking, Gemini Pro) — they already reason internally, so it's mostly overhead.
- Don't treat the default as a lock — every agent's model is movable from the Model dropdown.
Test it
- Open an agent → Settings → Identity → Model; confirm exactly one option is tagged
· default(Gemini 3 Flash). - Pick a high-reasoning model (e.g. GLM-5.2), Save, and run a prompt that needs reasoning.
- Switch the same agent to a standard model (e.g. Gemini 3.1 Pro), Save, and re-run the identical prompt — verify the reasoning depth drops, confirming depth follows the model.
- Toggle Think step by step on a cheap model and re-run — verify the answer now shows explicit step-by-step working.