The Meta Watermelon AI model, an unreleased frontier system from the company’s Superintelligence Labs, has reportedly caught up with OpenAI’s GPT-5.5 on a range of key benchmarks, according to a Benzinga report citing comments attributed to Meta’s AI chief Alexandr Wang. If the reporting holds, it would mark the first time in more than a year that a Meta model has been credibly described as matching an OpenAI frontier release, and it lands as Washington reportedly seeks to slow the pace at which OpenAI ships its own next-generation system.
Key takeaways
- Meta’s upcoming ‘Watermelon’ model reportedly matches OpenAI’s GPT-5.5 on key benchmarks, according to Benzinga citing Alexandr Wang.
- Business Insider separately reports that Meta’s AI leadership believes the company is ‘finally catching up’ to OpenAI after a difficult stretch.
- Mashable reports that the White House has asked OpenAI to limit the launch of its next model, a request that could reshape competitive timing.
- None of the sources published verified benchmark scores, release dates, parameter counts or pricing for the Watermelon model.
- The developments arrive amid intensifying scrutiny of frontier-model safety, national-security implications and compute economics.
What Benzinga’s report actually says about the Watermelon model
According to Benzinga, Meta’s forthcoming model — internally referenced by the codename ‘Watermelon’ — has reached parity with OpenAI’s GPT-5.5 on several key benchmarks. The outlet attributes that characterisation to Alexandr Wang, the former Scale AI chief executive who now leads Meta’s Superintelligence Labs unit. Benzinga’s snippet does not disclose specific benchmark scores, the evaluation suites used, the size of the model, or a firm launch window.
The absence of granular numbers is important. Frontier-model comparisons typically hinge on a mix of reasoning suites, coding evaluations, multilingual tests and safety scores, and ‘matching on key benchmarks’ can mean very different things depending on which subset is highlighted. Readers assessing the claim should therefore treat the parity framing as a directional signal from Meta’s own leadership rather than a verified independent result. We track shifting benchmark leaders in our AI models database.
Why Meta’s positioning has shifted
Business Insider, in a separate report, quotes Meta’s AI leadership as saying that the company is ‘finally catching up’ to OpenAI. That framing is notable because Meta spent much of 2024 and 2025 fielding criticism that its Llama family had fallen behind frontier closed models on the hardest reasoning and coding tasks, prompting the company to restructure its AI organisation and recruit heavily.
The Superintelligence Labs unit, now led by Wang, was set up to consolidate that push. Business Insider’s snippet frames the current moment as an inflection point in Meta’s turnaround, though it stops short of confirming that Watermelon has shipped or that external evaluators have replicated any parity claim. For teams weighing where to place their bets, our AI price-performance index tracks how each frontier release changes the competitive calculus.
The Washington angle: a slower cadence at OpenAI?
Mashable reports separately that the White House has asked OpenAI to limit the launch of its next model. The snippet does not disclose the reasoning behind the request, the exact model in question, or whether OpenAI has agreed to any restrictions. If confirmed, however, such a request would represent a striking intervention in commercial release timing, and it would arrive at a moment when Meta is claiming benchmark parity with GPT-5.5.
The interplay between the two reports is worth flagging as analysis rather than as reported fact: a slower OpenAI cadence, combined with a credible Meta challenger, could compress the perceived gap between the leading closed model and the leading open-weights contender faster than the benchmarks alone would suggest. Convly readers evaluating open-versus-closed trade-offs can consult our open vs closed AI cost study for the underlying economics.
How the reported claim compares to the current frontier
The table below summarises the state of what is publicly reported by the three sources cited in this article. It is intentionally sparse — anything not present in the snippets has been left blank rather than filled with speculation.
| Item | Meta ‘Watermelon’ | OpenAI GPT-5.5 |
|---|---|---|
| Status per sources | Upcoming; reportedly matches GPT-5.5 on key benchmarks (Benzinga) | Referenced as the benchmark of comparison; next model reportedly subject to White House launch request (Mashable) |
| Named executive | Alexandr Wang (Meta Superintelligence Labs) | Not specified in snippets |
| Benchmark scores disclosed | Not disclosed in sources | Not disclosed in sources |
| Release date | Not disclosed in sources | Not disclosed in sources |
| Distribution model | Not disclosed in sources | Not disclosed in sources |
The empty rows in that table are the point. For all the attention the story has drawn, the publicly reported evidence is narrower than the headlines suggest.
What the reports do not tell us
Several questions remain open based purely on the sources. First, none of the three outlets, based on the available snippets, discloses whether Watermelon will be released with open weights in keeping with Meta’s Llama tradition, or whether the Superintelligence Labs unit is pivoting toward a more closed distribution. That distinction matters enormously for developers planning their stack — an open-weights model that matches GPT-5.5 would carry different implications than a closed release, particularly for those weighing local deployment. Teams comparing hosted and on-prem economics can model the trade-offs with our self-hosting vs API calculator.
Second, the sources do not indicate any pricing information for either the Meta or the OpenAI system, so cost-per-token comparisons cannot be made responsibly from the reporting alone. Readers who want to compare current published rates for the models that are shipping today can use our AI API cost calculator.
Third, the reporting does not confirm which benchmarks Meta is referring to when it claims parity. That leaves open the possibility that the parity is concentrated in a subset of evaluations while other suites still favour OpenAI.
Why this matters for AI developers and buyers
Even accounting for the caveats, the combined picture from Benzinga, Business Insider and Mashable is competitively significant. Frontier-model competition has, for much of the past eighteen months, been defined by a perceived quality gap between OpenAI and everyone else. A credible parity claim from Meta — even one that has not yet been independently verified — reframes that narrative and puts pressure on procurement teams that had already standardised on a single vendor.
For engineering leaders, the practical takeaway is to hold procurement decisions loosely until Watermelon actually ships and is independently benchmarked. For platform teams that have been consolidating around a single frontier provider, the reporting is a reminder that vendor diversification remains a live consideration. Developers building agentic systems in particular may want to watch how each new model performs on the evaluations that matter for tool use and long-horizon tasks; our roundup of AI coding agents tracks how frontier changes propagate into the agent layer.
Reading the competitive signal without overreading it
It is worth treating the current news cycle with proportion. Meta’s leadership has an obvious incentive to characterise an unreleased model favourably, particularly after a period in which the company was described as trailing. OpenAI has an equally obvious incentive not to respond to unreleased competitors. The White House request reported by Mashable, whatever its ultimate scope, sits inside a broader debate about frontier-model governance that has been building for more than a year.
What can be said with confidence, based solely on the three sources cited here, is that Meta’s AI chief is publicly asserting parity with GPT-5.5 on key benchmarks; that Meta’s own AI leadership believes the catching-up phase is real; and that OpenAI’s next launch may be subject to policy pressure that its competitors do not face. Those three data points together justify closer attention to Meta over the coming weeks.
Frequently asked questions
What is Meta’s ‘Watermelon’ AI model? Based on the Benzinga report, Watermelon is the internal codename for an upcoming Meta AI model that Alexandr Wang has reportedly said matches OpenAI’s GPT-5.5 on key benchmarks. Specifics such as size, architecture and release date are not disclosed in the available reporting.
Has Watermelon been released? No. Benzinga describes it as an upcoming model, and none of the sources cited here confirm a launch date, pricing or distribution model.
Who is Alexandr Wang and why is he making this claim? Wang leads Meta’s Superintelligence Labs unit, the group responsible for the company’s most ambitious AI work. Business Insider separately reports that Meta’s AI leadership believes the company is finally catching up to OpenAI, which is the broader context for Wang’s parity remark.
What is the White House reportedly asking of OpenAI? Mashable reports that the White House has asked OpenAI to limit the launch of its next model. The available snippet does not specify the model or the nature of the requested limits.
Does this mean Meta has overtaken OpenAI? Not according to the sources. The reporting describes parity on key benchmarks rather than a decisive lead, and none of the outlets cited here has published independently verified scores. The claim should be treated as a directional signal, not a settled result.
The bottom line
The Meta Watermelon AI model story matters because it reframes the frontier as a genuine two-horse race again, at least on Meta’s own telling. Benzinga’s report, Business Insider’s characterisation of Meta’s momentum and Mashable’s reporting on the White House’s request to OpenAI each carry weight on their own, and together they describe an unusually fluid moment in the model race. Until Watermelon ships and independent evaluations arrive, however, the responsible posture is watchful patience rather than premature repositioning. Developers, procurement teams and policy watchers all have reason to pay closer attention to Meta’s next move — but they also have every reason to demand the numbers before rewriting their roadmaps.
Sources: news.google.com. Reported July 03, 2026.