What's new on LaunchDarkly

Guarded releases ensure minimum sample size

To ensure rollout and regression decisions are never made with zero or near-zero data, we’ve introduced a minimum sample size calculator. If a decision can’t be made by the time 50% of the audience has been exposed, we’ll notify you and automatically extend the rollout duration. If a decision is still not possible, users will be alerted and the rollout will be rolled back. You can then address the low sample size or choose to roll forward manually.

Support for multiple simultaneous experiments running on a single flag

You can now run multiple experiments concurrently on a single flag—one per targeting rule. This enables powerful optimization scenarios, letting customers experiment across distinct audience segments without waiting for one test to end before starting another. It’s ideal for teams looking to accelerate experimentation and rollout strategies across diverse user groups.

Other improvements

• Added the ability to restrict AI model configurations, so that some models are not available for members when they create new AI Config variations.

• Enhanced guarded rollout data visualizations with greater granularity.

• Added support for leaving comments on AI Config updates.

• When targeting is toggled off, flags now display the number of rules, targets, or prerequisites (if any).

• You can now view variation details by hovering over the diamonds in the flag list.

Bug fixes

• Fixed a display issue with experimentation layer traffic allocation in experiment create/edit modes.
• For Snowflake-native experiments, we now correctly show no results if Snowflake has no exposure data—previously, LaunchDarkly exposure counts were shown instead.
• Ensured experiment exposure data is always sorted by timestamp.
• Flags in experimentation layers can now have Data Export enabled.
• Fixed a horizontal scrolling bug on the AI Configs list.
• Resolved an issue with updating approvals when permissions were based on project tags.