Meta's chief AI officer told employees at an internal town hall that the company's next flagship model, codenamed Watermelon, has matched OpenAI's GPT‑5.5 on closely watched benchmarks. He did not name which ones, and neither company has confirmed the claim publicly.
Watermelon succeeds Muse Spark, Meta's April release, and is described as using an order of magnitude more training compute — a run reportedly happening on a new Ohio data-center cluster built specifically for this scale of training.
The catch: this is a single, anonymously sourced, internal statement with no published benchmark table and no third-party verification. Independent trackers currently place Muse Spark behind GPT‑5.5, Claude, and Gemini on the Intelligence Index — so Watermelon's real position won't be known until Meta actually ships it.
OpenAI's GPT‑5.6 lineup — Sol, Terra, and Luna — launched under a restriction requested by the US government, which asked that the most capable tier be limited to roughly twenty pre-cleared partner organizations before any wider release.
The company says Sol crossed the "high" risk threshold on its own cybersecurity benchmark, which triggered the review. OpenAI has publicly pushed back on the idea that government pre-clearance should become a permanent step for future launches, even while complying with it now.
The pattern echoes what happened to Anthropic's Fable 5 and Mythos 5 weeks earlier, when similar export-style controls briefly took both models offline entirely before access was restored to a vetted group.
Xbox CEO Asha Sharma told staff the division will cut roughly 3,200 roles over the coming fiscal year — 1,600 effective immediately, the rest tied to divesting four studios, including Compulsion Games and Double Fine, which will go independent.
The cuts sit inside a wider Microsoft reduction of 4,800 jobs company-wide. Sharma's memo pointed to five years of heavy spending against shrinking Xbox revenue, calling the current business "not healthy" rather than pinning the decision on generative AI automation.
Notably, no previously announced games are being cancelled as part of the reduction, and Minecraft and The Elder Scrolls are named as continued priority investments even as the division shrinks around them.
Cloudflare announced it will split AI bot traffic into three categories — Search, Agent, and Training — and will block Training and Agent crawlers by default on ad-supported pages for new and free-tier customers starting September 15.
Mixed-purpose crawlers that bundle search indexing with model training, including Googlebot, will be judged by the most restrictive applicable rule, meaning sites that block training could lose search visibility from crawlers that do both.
Cloudflare is also rebranding its Pay‑Per‑Crawl marketplace to Pay‑Per‑Use, launching with Ceramic.ai and You.com as partners who compensate publishers when their content actually shows up in an AI answer, not merely when it's fetched.
Plaid has held preliminary discussions with banks about a possible IPO, according to people familiar with the matter, though the company stresses no final decision has been made and it isn't racing toward a listing.
The bank-data connectivity company's valuation climbed back to $8 billion this year, still below its 2021 peak of $13.4 billion, after a leaner $6.1 billion round in 2025. Backers include Goldman Sachs, Citigroup, Mastercard, and Visa.
Fintech IPOs have mostly sat out a record $127 billion first half for US listings overall. A Plaid debut could land ahead of a possible Anthropic share sale reportedly under consideration for later this year.