Meta Avocado model delayed after weak tests

Big spending does not guarantee frontier-model readiness, and this delay highlights how hard it is to ship a competitive model at scale.

• –Internal benchmark gaps in reasoning, coding, and writing appear to be the immediate blocker.
• –A second delay weakens Meta’s timing advantage in a fast-moving model cycle.
• –If Meta pivots further from open releases to controlled access, developer sentiment could shift.
• –The episode reinforces that eval quality, not just compute budget, determines launch velocity.