Consensus answer
"Watermelon" is the reported internal codename for a next-generation frontier artificial intelligence model currently under development at Meta firstpost businessinsider. It is intended as the successor to "Avocado"—the internal codename for Meta’s Muse Spark model, which was released in April 2026 firstpost businessinsider.
According to reports from an internal company town hall held in early July 2026, Alexandr Wang, the head of Meta’s Superintelligence Labs, stated that the Watermelon model utilizes an order of magnitude more compute than its predecessor firstpost windowsreport.
Regarding its performance, the claim is that Watermelon has "caught up to" or reached parity with OpenAI's GPT-5.5 on certain internal, closely followed benchmarks americanbazaaronline benzinga businessinsider. However, several factors suggest caution in interpreting these reports:
- Unverified Claims: These performance assertions are based solely on Meta’s internal evaluations firstpost benzinga. No specific benchmark scores, methodology, or model cards have been released to the public, and the results have not been verified by independent third parties firstpost benzinga businessinsider.
- Development Status: The model is currently in the training phase and has not been released for public use or external testing firstpost businessinsider.
- Competitive Context: The AI landscape continues to evolve rapidly. Reports indicate that OpenAI has already provided a limited preview of a successor model, GPT-5.6, which is reportedly even more capable than GPT-5.5, though its availability is currently restricted cifrum windowsforum techtimes.
In summary, while Watermelon represents a significant investment by Meta in compute and infrastructure to compete at the frontier level, there is no evidence that it currently "beats" GPT-5.5. The existing claims of parity are preliminary, internal, and lack the independent validation required to determine its actual performance capabilities.
Differences
1 notable difference (1 contradiction) across 6 models.
The existence and status of the 'Watermelon' model and 'GPT-5.5'.
These models are internal/unreleased projects reported in industry news.
Gemini, Grok, Anthropic, Mistral
“is the internal codename for an upcoming, in-training artificial intelligence model”
These models do not exist or are not confirmed by credible sources.
DeepSeek, OpenAI
“There is no publicly known or confirmed "Watermelon model" from Meta”
How to verify: Check if the user is asking about a hypothetical scenario or a specific set of leaked reports from July 2026.
Best answerMistral
Sources
- 1 businessinsider.com vertexaisearch.cloud.google.com
- 2 aiweekly.co vertexaisearch.cloud.google.com
- 3 letsdatascience.com vertexaisearch.cloud.google.com
- 4 americanbazaaronline.com vertexaisearch.cloud.google.com
- 5 Introducing GPT-5.5 | OpenAI openai.com
- 6 Meta says its next-gen AI model 'Watermelon' uses 10x more compute, matches GPT-5.5 benchmarks – Firstpost firstpost.com
- 7 Meta AI chief says ‘Watermelon’ model has caught up to GPT-5.5 americanbazaaronline.com
- 8 Meta's Watermelon AI Model Matches OpenAI's GPT-5.5 on Key Benchmarks | Technobezz technobezz.com
- 9 Meta's Upcoming 'Watermelon' AI Model Matches OpenAI's GPT-5.5 on Key Benchmarks, Alexandr Wang Reportedl - Benzinga benzinga.com
- 10 Meta's Watermelon Matches GPT-5.5 Benchmarks | Let's Data Science letsdatascience.com
- 11 Meta Watermelon vs GPT-5.5: what is known so far today cifrum.kz
- 12 Meta Watermelon AI Claims GPT-5.5 Benchmark Catch-Up: Windows IT Impact | Windows Forum windowsforum.com
- 13 Meta Watermelon AI Claims GPT-5.5 Parity: Benchmarks Remain Unnamed and Unverified techtimes.com
- 14 Meta Claims Its Watermelon AI Model Has Caught Up With GPT-5.5 windowsreport.com
- 15 businessinsider.com
- 16 ai.meta.com
- 17 developers.openai.com
Cite this answer
consens.io. (2026-07-05). Consensus answer to "What is Meta’s Watermelon model and does it beat GPT-5.5?". Models consulted: OpenAI: GPT-5.4 mini, Mistral: Mistral Small 4, Anthropic Claude: Claude Haiku 4.5, Google Gemini: Gemini 3.1 Flash-Lite, DeepSeek: DeepSeek V4 Flash, Grok: Grok 4.20. Consensus model: Gemini. Sources: https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQHVjQA3zY0U0PcF9-nNYyjB2_QLCg6tIeo9yR6Q893gvwB2yZRuU6U2-VMaH-xgGktef3gNyFRskd_zA8TiaXCv9ohXYGLoqzlg_Zd3Q89iZ_TcwOIt4OzVGNjJ-QpL-khhR4iBBGuP4XsD1izIMo63CTtrR8W8NVtjSIOmcVbwg9IjbkyprYU=, https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQFTst3Kes15Gh5FhhgSLucGEbi4ojG_hYcHZxkF0tJaLgzwiimwBjAz1KmRfDpu6to8vGo68zibqJ1xohXmOyxta5AJxhME2HV-g7xDnNmejsDeeQ==, https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGVAc79P6B5sA8Dt3U2w7DvPZGzjeVY3iultBKjo1Uslv7Xgg2toAOmSr5ViHqDcF1vdK12fI28HTMvv9jYra52WzR-5t8FAhPUTw_N1Lea_TeOYfPXfqyEGa2l0LQek6nYBb809sDsWWO0GkCYjQg-QY1nSrU-y-qLSJA4GdO3VsAltxZCIKr8oAs=, https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQGasHchPrJubg1APAOBecRdZUoc6r_LI6ekh2YCvfrbwP-i9mn74X3qOL8N4tqwrg0eVc1-jz3qA2VQei7BP_6pbQVBD0gntji6Co5dg09mVqLUj-xtA-CAIKNapSDQ_X_z8eUvrqdIyiTqVfQWPg9dy72vrOhctpIMPMZLPGh__ikoeqEzp7Ovp0AaZX_knqQkJVObbIked-iXddtP6T45-wKjFcEE, https://openai.com/index/introducing-gpt-5-5/?utm_source=openai, https://www.firstpost.com/tech/meta-says-its-next-gen-ai-model-watermelon-uses-10x-more-compute-matches-gpt-5-5-benchmarks-14028469.html, https://americanbazaaronline.com/2026/07/03/meta-ai-chief-says-watermelon-model-has-caught-up-to-gpt-5-5-484022/, https://www.technobezz.com/news/metas-watermelon-ai-model-matches-openais-gpt-55-on-key-benchmarks, https://www.benzinga.com/markets/tech/26/07/60264651/metas-upcoming-watermelon-ai-model-matches-openais-gpt-5-5-on-key-benchmarks-alexandr-wang-reportedly-tells-employees, https://letsdatascience.com/news/metas-watermelon-matches-gpt-55-benchmarks-76a9460e, https://cifrum.kz/en/meta-watermelon-gpt-5-5-what-is-known/, https://windowsforum.com/threads/meta-watermelon-ai-claims-gpt-5-5-benchmark-catch-up-windows-it-impact.433806/, https://www.techtimes.com/articles/319723/20260704/meta-watermelon-ai-claims-gpt-55-parity-benchmarks-remain-unnamed-unverified.htm, https://windowsreport.com/meta-claims-its-watermelon-ai-model-has-caught-up-with-gpt-5-5/, https://www.businessinsider.com/meta-ai-model-catches-up-openai-gpt-5-says-2026-7, https://ai.meta.com/blog/introducing-muse-spark-msl/, https://developers.openai.com/api/docs/models/gpt-5.5 Retrieved from https://www.consens.io/s/what-is-metas-watermelon-model-and-does-it-beat-gpt-5-5-S7Oa801KGgp6c9By
Related questions
- GPT-6 Leak: Is OpenAI launching GPT-5.7 or GPT-6 in August 2026 with a 1.5M-token context window, a new 10T-scale pretraining foundation, and major agentic reasoning gains? Separate confirmed facts fr 5 models compared
- Gemini 3.5 Pro Leak: Is Google’s ‘Cappuccino’ AI model really launching on July 17 with a 2M-token context window, advanced coding, AI agents, and deeper reasoning? Separate confirmed facts from rumor 5 models compared
- Is Grok 4.5 better than GPT 5.5 or Opus 4.8? 6 models compared
- LongCat-2.0 Explained: The Meme-Cat Model That Secretly Topped OpenRouter as Owl Alpha 6 models compared
This page is a snapshot of an AI-generated consensus answer created with consens.io. AI models can make mistakes – verify important information using the sources above.
Thanks – this page has been reported for review.