Release notes··3 updates
3 updates shipped on April 9, 2026.
Update 01 of 03
Fast (Haiku) for quick questions, Intelligent (Sonnet) for deep analysis.
The chat composer now has a model selector. Fast is good for quick edits, navigating the platform, and short questions. Intelligent is the right choice for complex generation, deep analysis of failed runs, and anything where you want the agent to think harder. Your preference persists across sessions.
Update 02 of 03
Re-running an entire run no longer overwrites the original.
Previously, hitting "Re-run all" on a run reset the existing run and replaced its results. Now it creates a new run, leaving the original intact. You can compare them side by side instead of losing the old result.
Update 03 of 03
New organizations default to $0 with clear copy on what that means.
The spending limit input was confusing: a $0.00 placeholder looked like an actively set value when it wasn't, and new organizations defaulted to no limit at all. Now new organizations default to a $0 spending limit with enforcement enabled, so any on-demand usage requires an explicit opt-in. The placeholder reads "No limit set" with dynamic help text showing whether on-demand usage is currently capped.
Budget alerts are also on by default for all customers. You'll get a heads-up before you hit a limit, not after.
See how QA.tech agents test your product in a 30-minute demo – and leave with a plan to reclaim those hours.