Meta Delays Next AI Model Release to May

Meta Platforms has delayed the release of its next major artificial intelligence model, codenamed Avocado, from this month to at least May, according to sources familiar with the matter. The setback comes amid the company's aggressive push to compete in the fast-evolving AI landscape, where rivals have advanced more quickly.

Internal testing revealed that Avocado fell short in key areas such as reasoning, coding, and writing when benchmarked against leading models from competitors. The model outperformed Meta's previous offerings and Google's Gemini 2.5, released in March of last year, but it did not match the capabilities of Gemini 3.0, which launched in November. Sources indicated the delay could extend to June, reflecting the need for further refinement to meet Meta's high standards.

Meta's last truly frontier-pushing AI model was Llama 3, unveiled in 2024. The company followed with lighter versions of Llama 4 last year, optimized for devices with limited hardware. Plans for a more powerful successor, internally referred to as Behemoth, have not materialized amid the broader slowdown. This marks a rare admission of vulnerability for Meta, which has positioned itself as a key...