Update
GPT-6 Luna is the default model for new workspaces, research agent included
TL;DR
GPT-6 Luna is now the default model for a new workspace's orchestrator, specialists and research agent. The research agent's default was DeepSeek V4 Pro, and the preset still named GPT-5.6 Luna while the platform default had moved on.
Why Luna
- Price. $0.10 per million input tokens and $0.50 per million output, against $0.66 / $1.98 for DeepSeek V4 Pro, the previous research default.
- Per task. The measured median task costs 29 credits in model tokens on GPT-6 Luna, 60 on GPT-5.6 Luna and 163 on DeepSeek V4 Pro (see credits by model).
- The two settings had drifted apart. The platform default was GPT-6 Luna while the preset a new workspace is seeded with — and that the Agents page showed — still said GPT-5.6 Luna.
Who was moved, and who was not
- Moved: 36 workspaces whose agents were all exactly on the old untouched defaults (33 free, 3 internal). Each carries a record of what it was moved from, and an audit row.
- Not moved: every workspace where anyone had picked a model, every Bring-Your-Own-Model workspace, and 16 workspaces on older research defaults.
- Unchanged: the safety agent stays on Gemini 3.8 Flash.
Any agent's model can be changed on the Agents page; a stronger orchestrator with cheap specialists remains the usual split.
Frequently asked questions
- What is the default AI model in Agent Planners?
- GPT-6 Luna, for a new workspace's orchestrator, specialists and research agent; the safety agent stays on Gemini 3.8 Flash. Every agent's model can be changed on the Agents page.
- Was my workspace's model changed?
- Only if every agent was still on the untouched old default. A workspace where anyone had chosen a model, or that runs its own model (BYOM), was not touched.