AI BriefingMeta Testing Various Models Including Avocado 9B and Avocado Mango Agent
Key point
As Meta delays the launch of its next-generation AI **Avocado**, it is narrowing the performance gap through testing of various variant models and parallel use of **Gemini**.
Details
The launch of Meta's next-generation AI model Avocado has been delayed from the original March timeline to after May 2026. This follows internal test results showing that its performance falls short of the leading models from Google, OpenAI, and Anthropic.
Meta is currently experimenting with various Avocado variant models in parallel. The confirmed key configurations are as follows:
- Avocado 9B: A small-scale model with 9 billion parameters
- Avocado Mango: A multimodal Agent model capable of image generation
- Avocado TOMM: A model based on 'Tool of many models'
- Avocado Thinking 5.6: The latest version of the Thinking model
- Paricado: A text-only conversational model
To close the performance gap, Meta is supplementing its capabilities with external models by temporarily licensing Google Gemini technology or conducting A/B tests that route some requests to Gemini models.
Also, in line with CEO Zuckerberg's policy of pursuing superintelligence, Avocado is expected to break from the existing open-source tradition and launch as a proprietary model. This is a move that contrasts with the Llama era.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.