Three changes to how Account Qualification, Account Research, and Contact Research agents run. They all point in the same direction: you get the best frontier model for each task, at the best price, by default.
Three changes to how your research and qualification agents run. The model dropdown is gone. You pick the effort level, we route each run to the strongest frontier model available for that level. When better models ship, your agents inherit them automatically.
OpenAI, Anthropic, and Google ship new frontier models every few weeks. Capabilities shift. Prices shift. Which model is best for account research this week is not the same as which one is best for contact qualification, and both change again next month. Your team should not have to track any of that.
So the model dropdown is gone on these three agents. You now pick Low, Medium, or Advanced effort, and we route each run to the strongest frontier model for that level of reasoning, at the best price we can secure. When better models ship, your agents inherit them automatically. No migration. No retuning. No settings to update.
Account Qualification, Account Research, and Contact Research agents can now run at Standard or Fast. Standard is the default and covers most use cases. Fast doubles the credit cost and gives you priority processing for moments where timing is the whole point: inbound lead qualification, post-event follow-up, day-of outbound on a fresh signal.
The model now powering Low is a frontier reasoning model, significantly more capable than what we ran before. In our testing, it handles account research, qualification, and contact-level tasks at a quality level we could previously only get from much more expensive models. The extra 0.2 credits reflects that jump in capability. Low is still the cheapest way to run any agent in Evergrowth.
You don't need to do anything. We re-routed every existing agent to the most efficient frontier model behind the scenes, with no change to your output structure or schema.
The other handful, the ones doing complex multi-step reasoning where instructions can't fully cover the path, will perform best on Medium or Advanced. Your CSM will flag the specific ones worth retuning.
This release covers three agent types. Every other agent runs as it did before, for now.
Coming next. Same framework rolling across the rest of the agent types over the coming weeks, starting with Contact Qualification and Play Agents.
The new Effort and Speed controls are visible in two places. The model dropdown is gone on these three agents.
If your agent output style is locked to a specific phrasing or structure, run a few sample runs on the new Low before you trust the outputs at volume. The quality bar is higher, but the phrasing can shift.
For agents doing complex multi-step reasoning where the steps can't be fully spelled out, Medium or Advanced will outperform Low. Your CSM can walk through any agent you want to retune.
Contact Qualification and Play Agents are next, so you can dial speed on outbound sequences where timing matters most. We'll roll the rest out over the coming weeks.