Meta stops using fruit codenames for AI models after "Watermelon," the internal name for Wonder Full
Meta has internally decided to stop naming its AI models after fruit, a person familiar with the matter told Business Insider. "Watermelon" was the codename for Wonder Full, which Meta said is its first cutting-edge "frontier" model.
Meta has internally decided to stop naming its AI models after fruit, according to a person familiar with the matter cited by Business Insider (reporter Charles Rollet, published November 4, 2026). Meta declined to comment.
What the report says
The person described the change as largely a branding decision, because fruit-based codenames "don't chime well" with Meta's lofty AI ambitions.
"Watermelon" was the internal codename for Meta's latest AI model, Wonder Full. Meta said Wonder Full is its first cutting-edge "frontier" model, and it teased the launch by posting photos of the fruit.
Meta's earlier models also carried fruit names:
- Avocado: the previous language model
- Mango: the previous image model
- Watermelon: Wonder Full
Executives on the record
Meta CEO Mark Zuckerberg and AI chief Alexandr Wang joked on the Cheeky Pint podcast last month that they had run out of fruit names. "We're out of fruits," Zuckerberg said. "So we're going to have to switch to a new convention."
Wang added: "I think the only fruit that is technically larger is the durian, and you don't want to go to the durian."
Wider naming context
Business Insider notes that Meta is not the only company to use grocery-store names. The Information reported in 2025 that OpenAI used "Garlic" as an internal model codename. Google released its "Nano Banana" image generator in 2025.
The outlet also says it has become trendy to give more powerful models awe-inspiring names, citing Anthropic's Mythos and OpenAI's Astra.
Model details
The report does not cover Wonder Full's context window, parameter count, benchmark scores or training cutoff. Pricing not yet disclosed. Meta has not announced a replacement naming convention.
What this means
This is a codename and branding change, not a new model release. The only model named in the report, Wonder Full, was already described by Meta as its first frontier model, and Watermelon was its internal label.
For anyone tracking Meta's models, internal fruit names (Avocado, Mango, Watermelon) will not carry forward to future releases, according to the person familiar with the matter. Zuckerberg's own comments on the podcast point the same way: he said Meta will "have to switch to a new convention." Future internal codenames from Meta will follow a different scheme, though the report does not say what it will be.
Related Articles
Replicate advises starting with MiniMax H3 Highspeed at $0.06 per second, using H3 for 2K
Replicate recommends starting with minimax/h3-highspeed, dropping to minimax/h3-lite for large batches, and moving up to minimax/h3 for 2K output or difficult shots. A 10-second clip at 768P costs $0.60 on Highspeed, $0.40 on Lite and $0.80 on H3.
Revolut cuts AI coding inference costs roughly 90% by making self-hosted GLM models its default
Revolut says it cut AI coding inference costs by roughly 90% in less than a year by making self-hosted GLM models its default over commercial tools like Claude Code. The fintech serves GLM-5.3 on 56 Nvidia B200 GPUs to roughly 2,000 engineers, according to infrastructure lead David Abitbol.
OpenAI adds task router to Dots to cut compute as Musk calls Grok Bot costs 'unsustainably high'
OpenAI announced an automatic router that sends Dots tasks to the smaller GPT-6.1 Terra model rather than the larger GPT-6 Astra, to reduce compute use. Elon Musk said this week that Grok Bot's running costs are "unsustainably high relative to revenue." Fortune reports that compute supply is the main constraint on personal AI agents.
Google Genie 4 points to new uses for world models in robotics and AI testing
Google's updated Genie 4 points to new uses for world models in robotics and AI testing, according to AI Business. The outlet says enterprises must weigh that potential against accuracy gaps.
Comments
Loading...