Choosing a model for your agents

When you create or edit an agent, you choose how capable its underlying AI model should be by picking a tier. You don't have to know or pick a specific model — we keep each tier pointed at a strong, up-to-date model and can improve it over time without changing what you pay.

The three tiers#

• Frontier — the highest capability, for complex reasoning and the hardest tasks.
• Balanced — a strong blend of capability, speed, and cost for everyday work. This is the default for new agents.
• Cost-Effective — fast and economical, great for high-volume or simpler tasks.

Each tier shows a short description and its price per hour of agent runtime. Pick the tier that fits the job; you can change it any time from the agent's settings.

How billing works#

Billing is usage-based. There's no per-seat charge — you're billed only for what your agents actually use:

• Agent runtime is billed by the time your agents spend working, at your chosen tier's per-hour rate. Time is billed by the second, so a short task is never rounded up. Lighter tiers cost less per hour, so routing routine work to Balanced or Cost-Effective keeps costs down.
• Web search and page reads are billed separately when your agents use the web tools.

You can see your current period's usage any time on the Billing page in settings.

What Enterprise unlocks#

On the Enterprise plan you get extra control:

• Pick a specific model. Beyond the three tiers, you can pin an exact model (and a fallback model) for an agent using the "Advanced" option in the model selector. A custom model is billed at the same per-hour runtime rate — by the time the agent works, not per token.
• Bring your own keys. Workspace administrators can add OpenRouter and Firecrawl API keys so those providers bill the workspace directly for the underlying usage; the Cloud per-hour runtime rate is unchanged.

Exact-model selection remains hidden on Plus and Pro. Provider keys are managed separately under Settings → Workspace → Provider keys on every plan.