The Muse variant consumes ten times more compute than Avocado.
SCOOP: Alexandr Wang says Meta's upcoming AI model - codenamed Watermelon - has caught up to OpenAI's GPT-5.5.
"Watermelon uses an order of magnitude more compute than Avocado," he said in an internal meeting, referencing Meta's previous AI model.
https://www.businessinsider.com/meta-ai-model-catches-up-openai-gpt-5-says-2026-7
Some users are excited that Meta's Watermelon model could shift the AI race by matching GPT-5.5 while still training, while many others dismiss the claims as hype and criticize Meta for lagging competitors on benchmarks.
No Digg Deeper questions have been answered for this story yet.
According to Alexandr Wang, Meta’s next frontier model, codenamed Watermelon, has already caught up with OpenAI’s GPT-5.5 on internal benchmarks.
The model is still training.
One interesting detail: Watermelon reportedly uses an order of magnitude more compute than Avocado (Meta’s internal codename for Muse Spark).
Via Business Insider
Meta is expected to release an Opus model in the very near future.
Reportedly, Mark Zuckerberg’s disappointment with Agentic AI was directed at the field of agentic AI in general rather than at Meta specifically. What exactly he meant by that remains unclear. Personally, I don’t share that disappointment at all. Quite the opposite. I believe 2026 will be the year of Agentic AI, and I see very little reason to doubt that.
What interests me even more, however, is Meta’s actual objective. For the consumer market, its model is already good enough. Of course, updates are always welcome, and better models are never a bad thing. But in the business and enterprise space, Meta’s Spark is still significantly behind.
The only truly compelling agentic use case I currently see for Meta is in commerce. Imagine AI agents on Instagram that can purchase products directly, complete the entire transaction autonomously, and even handle the payment themselves. That would ultimately generate additional revenue for Meta.
That’s what I’d really like to understand.
now I lowkey want Meta to ship a model better than GPT 5.5 just to see Hensen crash out. This would be the ultimate black night of the soul. Falling behind Anthropic is… at least a dignified loss. But Zuck and Wang?!
Doubt this
One of the new Muse variants is code named Watermelon, and they are claiming that it is on par with GPT-5.5.
SCOOP: Alexandr Wang says Meta's upcoming AI model - codenamed Watermelon - has caught up to OpenAI's GPT-5.5.
"Watermelon uses an order of magnitude more compute than Avocado," he said in an internal meeting, referencing Meta's previous AI model.
https://www.businessinsider.com/meta-ai-model-catches-up-openai-gpt-5-says-2026-7
Just wait until you see the amount of compute used for the great pumpkin
SCOOP: Alexandr Wang says Meta's upcoming AI model - codenamed Watermelon - has caught up to OpenAI's GPT-5.5.
"Watermelon uses an order of magnitude more compute than Avocado," he said in an internal meeting, referencing Meta's previous AI model.
https://www.businessinsider.com/meta-ai-model-catches-up-openai-gpt-5-says-2026-7
@basedjensen > Wang > open sourcing near-frontier brother, 'fraid the best I can do is "lukewarm attempt to undercut on price"
https://www.businessinsider.com/meta-ai-model-catches-up-openai-gpt-5-says-2026-7
According to Alexandr Wang, Meta’s next frontier model, codenamed Watermelon, has already caught up with OpenAI’s GPT-5.5 on internal benchmarks.
The model is still training.
One interesting detail: Watermelon reportedly uses an order of magnitude more compute than Avocado (Meta’s internal codename for Muse Spark).
Via Business Insider
@teortaxesTex Rofl yea I would be surprised if they do catch up any time soon (<1 year) but if they do and open-sourced it I will gladly eat crow
@xeophon I think 5.6 is a relatively stranghtforward uplift of 5.5 if you can make 5.5 and have Meta's compute stash you can make 5.6 I'm not asking for miracles here
@airkatakana Unless they release it in less than a week I don't think they will be competing with GPT-5.5 anyway.
Open source it and we'll care even more!
We need Meta back in the American open source champion category.
Come back to the side of the light! We miss you.
According to Alexandr Wang, Meta’s next frontier model, codenamed Watermelon, has already caught up with OpenAI’s GPT-5.5 on internal benchmarks.
The model is still training.
One interesting detail: Watermelon reportedly uses an order of magnitude more compute than Avocado (Meta’s internal codename for Muse Spark).
Via Business Insider
@teortaxesTex 5.5 isn’t enough with 5.6 being around the corner. Unless they finally start selling tokens and make those ridiculous cheap
@AndrewCurran_ (x) doubt
@kimmonismus Oh, I'm surprised 😯. Are they playing the Anthropic game here?
@teortaxesTex 🤣🤣🤣🤣
@Balu0X yeah, i wonder too
@sin4ch shots fired :D
@AndrewCurran_ This has gotten very very interesting. Shocked that this news comes after market close, when the original reporting was during market hours. Makes me wonder….
@kimmonismus watermelon catching up while still training is either very good news or a sign that gpt 5.5 is not the bar people think it is
@AndrewCurran_ Just like every Chinese model these days is “on par with Opus”
@AndrewCurran_ So no open weights right
