- Stronger multi-step workflow completion and fewer tool calls on coding tasks.
- Better ability to gather context, spot plan gaps, and produce final deliverables.
- Improved calibration around irreversible actions and stronger resistance to adversarial inputs and prompt injections.
- Available through Muse Code (interactive coding workflows) and the Meta Model API (for embedding into apps).
- Meta did not publish an open-weights release for 1.3 and says open-weights are planned later.
- Public announcement lacks a complete pricing table, regional rollout schedule, and local-download option.
- Designed to improve coding tasks: debugging, refactoring, test generation, implementation planning.
- Aims to reduce token/tool call costs on long tasks and make agents more reliable for real-world actions.
- Safety and adversarial-resistance claims are highlighted for agent/automation deployments.
Useful?