Update
GPT-6.1 Terra is the default model
TL;DR
GPT-6.1 Terra is now the default model for a new workspace's orchestrator and specialists. It replaced Claude Sonnet 5.5 at half the price per token, and it read our largest planning prompt as about 105K tokens against Sonnet's 180K.
What we measured
- Planning. On 7 goals across 26 platforms, Terra and Sonnet 5.5 both planned on every expected platform with no lint findings. Terra's plans took longer on the largest prompt: 41 seconds against 29.
- Tool choice. On 32 tasks that find a tool on demand, Terra chose the right tool 31 times; Sonnet 5.5, Haiku 5.5 and GPT-6 Luna 32 times.
- Quick checks. Claude Haiku 5.5 reviewed planted plan problems 24 of 24 times in about 2 seconds, so it replaced Gemini 3.8 Flash at about a seventh of the price.
These are tests on our own tasks, not production averages on Terra yet. Any agent's model can be changed on the Agents page.
Frequently asked questions
- What is the default AI model in Agent Planners?
- GPT-6.1 Terra, for a new workspace's orchestrator and specialists; quick internal checks run on Claude Haiku 5.5. Every agent's model can be changed on the Agents page.
- Was my workspace's model changed?
- A workspace with no stored models follows the default. One where anyone chose a model, or that runs its own (BYOM), is not touched.