Meta delays 'Avocado' AI model release after internal testing showed it lagged Google/OpenAI/Anthropic

TL;DR

Financial-news account FirstSquawk reported on 2026-03-12 that Meta has delayed the release of its new AI model “Avocado” after internal testing revealed it lagged behind rival models from Google, OpenAI, and Anthropic in reasoning, coding, and writing performance. No primary source from Meta is linked; the tweet itself is the artifact. Filed as an industry-news signal — Sam’s accompanying quip (“only way to fix this is to 10x everyones salary even further”) refers to Meta’s widely-reported 2025-2026 AI talent-acquisition push.

Key claims

Meta delayed the release of a new AI model codenamed “Avocado” [tweet body].
The delay was prompted by internal testing showing Avocado lagged Google, OpenAI, and Anthropic frontier models on reasoning, coding, and writing [tweet body].
The post is from FirstSquawk, a financial-news squawk service; no direct Meta source is cited in the tweet [tweet body / @FirstSquawk bio].

Method

N/A — this is a one-line news squawk, not a research artifact. No methodology, evaluation details, or benchmark numbers are provided; the claim of “lagging on reasoning, coding, writing” is unsourced inside the tweet.

Results

N/A. No numbers, no evals, no benchmark names.

Why it’s interesting

Pairs with the other recent talent/lab-formation industry signals on the wiki: Saining Xie joining AMI Labs (LeCun's new world-model lab) — announcement (Saining Xie joining LeCun’s new AMI Labs) and the broader frontier-lab reorganization narrative. It also sits next to On the Slow Death of Scaling as data points worth tracking on whether scale + talent spend is still translating to frontier capability — Avocado’s reported under-performance despite Meta’s hiring spree is the kind of anomaly that piece argues about. Treat the specific claim with skepticism: it’s a third-party squawk with no linked source.