Editorial illustration for Meta Tests GPT-5.5 With 'Watermelon' Model
Meta Tests GPT-5.5 With 'Watermelon' Model
Meta's Muse Spark model launched in April to a shrug. Industry read: usable, but nowhere near frontier. Three months on, chief AI officer Alexandr Wang says the follow-up has closed that gap entirely.
The model goes by the internal codename "Watermelon," and Wang is telling people it performs on par with OpenAI's GPT-5.5. That's a big jump for a company whose last public release drew the verdict "not great, but back in the game." Wang has also teased an upcoming coding model he compares to Anthropic's Opus tier, Anthropic's top-of-line coding product and a benchmark rivals treat as the one to beat.
None of this happens in a vacuum. OpenAI isn't pausing development while Meta trains its next release, and GPT-5.5 itself won't be the finish line for long. But Meta has sunk roughly $145 billion into its AI buildout, and until now the return on that spending has been hard to see in the actual products. Wang's claims, if they hold up once Watermelon ships, would mark the first sign that the money is translating into something the rest of the field has to reckon with.
When Meta's Muse Spark landed in April, the read was "not great, but back in the game." Three months later, Alexandr Wang says its successor is already running even with OpenAI's GPT-5.5. The frontier isn't standing still while ‘Watermelon’ trains, but if Wang's claims and tease of an Opus-level coder hold, Meta's $145B AI spend may finally be buying results the rest of the field pays attention to.
Why this matters
Codenames like "Watermelon" are the closest thing we get to a real signal before a launch, and the fact that Meta is reportedly benchmarking it directly against GPT-5.5 tells us where the company thinks it stands. For developers building on Llama or considering it, that's worth watching closely: a head-to-head test against OpenAI's frontier model suggests Meta believes it's closer to parity than its last few releases implied. For founders, this is another reminder that the model you build around today could be outclassed in a quarter, so architecture choices that assume long-term stability in any one provider are riskier than they look.
Researchers should treat "Watermelon" as unconfirmed until Meta actually ships something with real benchmarks attached, not marketing language. We'd caution against reading too much into a leaked codename, but the pattern of Meta testing against OpenAI's newest release, rather than its own prior models, is the part worth tracking as this plays out over the coming weeks.
Related Reading
Browse all OpenAI news →
LLMs & Generative AI ChatGPT's 'Nerdy' tweak rewards goblin metaphors in answers, study finds
LLMs & Generative AI Google tests visual 'magazine-style' UI for Gemini 3 Pro users
LLMs & Generative AI AI Engineers Face Rising Costs, Need New Strategies for Efficiency
Related Topic OpenAI's GPT Transcribe Cuts Error Rate to 3.31%, Improving on GPT-4o
Open Source
Related Topic OpenAI Calls Apple's Lawsuit 'Rotten to Its Core' in Rebuttal
Policy & Regulation
From Archives GPT-4o, Gemini, and Claude Vision Can Reason Over Visual Details
From Archives AI Search Agents Struggle With Ambiguous Queries, Study Finds
From Archives Square's ChatGPT integration charges restaurants 6% fee for pickup orders
From Archives OpenAI engineers say they halved inference costs for guest ChatGPT users
Common Questions Answered
What is Meta's 'Watermelon' model and how does it compare to the Muse Spark model?
Watermelon is Meta's internal codename for a follow-up model that reportedly performs on par with OpenAI's GPT-5.5, representing a significant improvement over the Muse Spark model that launched in April. While Muse Spark was considered usable but not frontier-level, Meta's chief AI officer Alexandr Wang claims Watermelon has closed that performance gap entirely.
Why is Meta benchmarking the Watermelon model directly against GPT-5.5?
Meta's direct benchmarking against OpenAI's GPT-5.5 signals where the company believes it stands competitively in the AI market. This comparison suggests Meta thinks it has achieved closer parity with frontier models than its previous releases implied, which is significant for developers considering building on Llama or evaluating Meta's AI capabilities.
What upcoming AI product has Alexandr Wang teased alongside the Watermelon model?
Alexandr Wang has teased an upcoming coding model alongside the Watermelon model announcement. While specific details about this coding model are limited in the article, its development indicates Meta's continued expansion into specialized AI applications beyond general-purpose models.
What does the industry's reaction to Meta's Muse Spark model reveal about the company's previous AI position?
The industry's lukewarm response to Muse Spark—described as usable but nowhere near frontier capabilities—indicated that Meta had fallen behind leading AI companies in model performance. This verdict of 'not great, but back in the game' showed Meta was attempting to catch up but hadn't yet achieved competitive parity with frontier models like GPT-5.5.
Further Reading
- Alexandr Wang says Meta's finally catching up to OpenAI - Business Insider
- Meta's Upcoming 'Watermelon' AI Model Draws Even with OpenAI's GPT-5.5 - TechStrong
- Meta Watermelon AI Claims GPT-5.5 Parity - Benchmarks Remain Unnamed, Unverified - Tech Times
- Meta's Watermelon Model Matches GPT-5.5 on Benchmarks - LinkedIn (AI Insiders News)
- According to Alexandr Wang, Meta's next frontier model, codenamed Watermelon, has already caught up with OpenAI's GPT-5.5 - X (Twitter)
