Skip to main content
Choose your agent’s AI model from the model dropdown at the top of the Agent tab.

​
Choosing a model

Open the model dropdown at the top of the Agent tab. The fastest way to choose is one of the options at the top:
  • Auto: Gumloop picks a model per message instead of pinning one. New agents normally default to Auto; Enterprise organizations and accounts where Auto is restricted default to Balanced instead. Organization default-agent settings can select another model.
  • Balanced: a preset for everyday work, currently Grok 4.7.
  • Smartest: a preset for complex reasoning and multi-step tasks.
Each preset maps to a model that Gumloop keeps up to date, so its model can change over time. Enterprise organizations can customize preset models and runtime settings; see AI model governance & configuration.
On Enterprise plans, admins can restrict which models a Custom Role may use. Restricted models can appear in the picker marked Restricted by Admin, but you cannot select them. Where available, you can request access.
To pick a specific model instead, search by name or browse by provider: Anthropic, OpenAI, Google, DeepSeek, MiniMax, Z.ai, and more.
Auto suits most agents, especially when message difficulty varies. Choose Balanced or Smartest to follow a preset, or a specific model when you want to pin the agent to that model.

​
Auto: let Gumloop choose the model

Auto is Gumloop’s model router, not a model from a provider. Select it once, and Gumloop picks which model runs each message. Your initial selection depends on your organization’s defaults and model restrictions.
Model dropdown with Auto selected, described as an advanced model router powered by Gumloop Chew

Auto sits at the top of the model dropdown on the Agent tab.

​
How Auto picks a model

1

It reads your message first

Auto considers the task’s difficulty, required tools, and how the answer can be checked. It aims to choose a lower-cost model that can handle the task. While it decides, the chat shows Choosing the best model.
2

It runs on a real model, and tells you which

The chat names the model it uses, for example Using Grok 4.6. The same model appears in the chat’s model list and credit breakdown.
3

It can move up mid-run

Auto can switch to a stronger tier if the task needs it. The chat shows Switching to …, and the stronger tier stays in use for the rest of the run.
You can ask Auto to prioritize capability or cost. To require a specific model, select it in the model picker.

​
The tiers Auto chooses from

Auto chooses among five tiers. Their models can change over time; the chat shows the model that actually ran. The two lowest-cost tiers are not used for these requests:
  • Anything that writes to the real world (send, post, reply, create, update, delete, or invite) runs on Grok 4.6’s tier or above, including bulk “do this to each of these” sweeps.
  • Turns that use other agents (sub-agents, delegations) run on Grok 4.6’s tier or above, because those tools aren’t available on the cheaper tiers.
  • Messages carrying images or other non-text attachments stay on Grok 4.6’s tier, since routing only reads text and can’t judge what it can’t see.

​
Where Auto can be used

​
Auto and restricted models

On Enterprise plans, Auto follows your organization’s model rules.
  • Auto follows both organization-wide model restrictions and Custom Role allow-lists, including when it switches models. If no permitted model is available, the run stops.
  • Admins can restrict Auto itself. An agent already saved on Auto then uses a permitted model without routing.
  • Claude Fable models must be selected explicitly; Auto never chooses them.

​
Auto FAQ

Use gumloop-chew as the agent’s model_name, or --model gumloop-chew in the CLI. To request a routing decision without running the model, use Route model. Direct chat-completion calls require a specific model, not Auto.
No. Auto is Gumloop’s model router. When you select Auto, Gumloop chooses a provider model for each message rather than using one fixed model for every message.
The chat names the model Auto runs with Using … and shows Switching to … if it changes models mid-run. The chat’s model list and credit breakdown record the models that actually ran.
Auto follows your organization’s model restrictions and Custom Role allow-lists, including when it switches models or uses a fallback. If no permitted model is available, the run stops rather than bypassing the restriction.
Yes. Admins can restrict Auto itself through organization model restrictions or Custom Role allow-lists. An agent already saved on Auto then uses a permitted model without routing. See AI model governance & configuration.
You’re charged for the tokens of whichever model actually ran. The routing decision itself is not billed. See Credits for model rates and other agent charges.
Auto falls back to its baseline model, currently Grok 4.6, when it cannot make a routing decision. Your organization’s model restrictions still apply.
Choose Auto when you want Gumloop to select a model for each message, especially when task difficulty varies. Choose Balanced or Smartest to follow a preset, or select a specific model to pin your agent to that model. Preset models can change over time.

​
Reading the model card

Selecting or hovering a model opens a detail card so you can compare options at a glance:
  • Description: what the model is best at.
  • Speed and Intelligence: relative ratings to help compare models. Ratings can change as benchmarks are updated.
  • Provider and Context: who makes the model and how much it can read at once (in tokens and approximate words).
  • Tool Calling and Vision: capability checks.
  • Badges: extra labels that call out how a model is served. Models marked US-provider hosted are served by US providers under Zero Data Retention (ZDR) policies, and models marked 30-day data retention are not covered by ZDR.
For agents that read images or screenshots, pick one with Vision.

​
What do the lifecycle labels in the model picker mean?

Some models in the picker show a lifecycle label in place of their context window: If the provider retires an agent’s saved model, or your admin restricts it, the line under the model picker says so and names the model that runs instead:
  • Current model is no longer available - using a replacement model, when the provider has retired the saved model.
  • Current model is restricted - using a permitted model, for a model your admin restricted. A Request access link sits next to it.
If no permitted model can run, the line names no replacement.

​
Available models

These are the models you can choose for an agent, grouped by provider. Every one supports tool calling, so it can use your connected services and data. Models marked Vision can also read images and screenshots. Served through is the inference provider Gumloop sends the request to, which also decides which BYOK key can cover the model. New models are added continually, so the in-product picker is always the most up-to-date list.
Some older models are hidden from new selection but remain available to agents and presets already using them. All Qwen models, including Qwen3.8 Max, are hidden this way, which is why searching the picker for Qwen returns No matching models. Agents already set to Qwen3.8 Max keep running on it.

​
Where open models are served

Use the Served through column to check which provider key covers a model. A model routed through OpenRouter can be hosted by Fireworks, but a Fireworks API key does not cover that OpenRouter request. For provider-key setup, see Bring your own key.

​
Models with 30-day data retention

The Claude Fable family has a specific 30-day data retention policy. Anthropic keeps prompts and outputs sent to Claude Fable models for 30 days to check for misuse, then deletes them. Anthropic does not train on that data. This exception does not apply to other Claude models. The family carries a 30-day data retention badge, and:
  • They are off by default. An organization admin has to explicitly allow non-zero data retention models before anyone can select one. See AI Model Governance.
  • They are never chosen for you. They are excluded from the Balanced / Smartest presets, from Auto, and from fallback models. The only way one runs is if you pick it.
  • If the acknowledgement is later revoked, they become unavailable immediately and agents set to them fail until you switch models.

​
Thinking controls

Most modern models think before they answer. Open the Model dropdown in the Agent tab and choose Reasoning to adjust its reasoning controls. The choices depend on the selected model: Gumloop only shows the controls that model supports. Other tuning parameters remain under Settings > Advanced AI. There are three shapes:
  • Effort levels: pick how hard the model thinks, from low up to max. More effort usually means a better answer, more thinking tokens, and a higher cost.
  • On or off: turn thinking off for speed and cost on models that allow it.
  • A thinking budget: give the model a token allowance to think inside.
The table also includes controls for models still used by saved agents.
Your choices are saved per provider, so switching models keeps the settings you picked for each one. If a model does not support a level you saved, Gumloop uses the closest level it does support rather than failing the run.
On Gemini 3 and Gemini 4 Argon, the thinking level is the only tuning control offered. Temperature and the other randomness settings are not shown.

​
How do Fast and Ultrafast modes work?

Speed changes a supported model’s response speed and token pricing without selecting a different model. Standard is the default. In the agent’s right-hand configuration panel, open the model dropdown in the Agent tab and select Speed to see the selected model’s choices. The corresponding API setting is mode. Faster modes have higher token rates. Reasoning and Context window are separate settings. Only supported modes appear in the picker. Saved legacy models may also offer Fast; check their Speed controls. If you switch to a model that doesn’t support the saved speed, Gumloop drops to the highest speed that model supports. For example, Ultrafast becomes Fast on Claude 5.5 Opus. The picker, the request, and billing all use that speed. See Credits for speed-mode rates and how the mode actually served affects billing.

​
What are decision models?

Decision models classify and score information through the Decisions API. Gumloop supports Jev 1.14 and Jev 1.13 from TypeSafe (typesafe/jev-1.14 and typesafe/jev-1.13), served through OpenRouter, and GPT-6 Luna from OpenAI (gpt-6-luna), served directly through OpenAI. Jev is decisions-only. GPT-6 Luna supports both decisions and chat under the same model ID. Give the model some data and a question with defined answers. It returns a decision and probabilities you can use in code, such as whether a ticket is urgent or which team should handle it. Use Create decision to call either model through /decisions. For a chat reply from GPT-6 Luna, use /chat/completions instead. Agents can also use decisions in Computer scripts when permitted; see Code sandbox & secrets. Check List models for current availability and decision support. Your model permissions still apply.

​
How agents are charged

In agents, model cost is token-based and variable. You are charged for the tokens each message uses, which depends on the model, the length of the conversation, and the tools available. There are no fixed per-message tiers. Model calls bill at cost, converted to credits at $0.005 per credit, and each run also carries Compute and an Orchestration Fee. Select the credit count in any chat’s header to see its exact usage, and see Credits for the full breakdown.

​
Bring your own key (BYOK)

Provide your own provider API key to cut model costs. On usage-based pricing, eligible BYOK model calls use 0 model credits, including eligible image-generation and transcription calls. Agent chats still have an orchestration fee: normally 16% instead of 8%, calculated on what the run would have been worth. Older organizations may use a different pricing policy, with a discount rather than a full model-credit waiver. See Credits.
A provider key covers only models served through that provider. Check the Served through column above; for open models, see Where open models are served.

Setting up a key

Requires a Pro plan or higher and your own OpenAI, Anthropic, Google AI, Perplexity, SpaceXAI, or Fireworks AI account.
Add a key on your personal AI Model Keys page (Customize > AI Model Keys in the left sidebar) so your own calls route through it, or add a shared team key so the whole team can use it without managing individual keys.Pro users cannot set keys at the organization level.