The trash talk between rival AI labs is getting more interesting as we go deeper into the AI revolution.
Right after OpenAI wrapped its livestream introducing GPT-5.6 Sol, Terra, and Luna, Anthropic’s ClaudeDevs account posted a two-line update: five-hour and weekly rate limits had been reset for every user. No context, no explanation, just the announcement. Tibo Sottiaux, the OpenAI executive who had spent the previous day on stage walking through Sol’s benchmark wins, quote-tweeted it with three words: “I smell fear.”
The post has since pulled in over 44,000 likes and 435,000 views, which is a lot of engagement for a comment that’s essentially a jab wrapped in a joke. Sottiaux isn’t some junior social media hire either — he’s the Head of ChatGPT and Codex at OpenAI, the executive who led the GPT-5.6 launch livestream and has become one of the more visible faces of OpenAI’s product side over the past year. When someone at that level takes a direct shot at a competitor within a day of a major release, it tends to travel.
Why The Timing Reads The Way It Does
Anthropic’s rate limits have been a sore point for its user base for a while now. Back in July 2025, the company introduced weekly limits on Pro and Max plans after some users reportedly burned through tens of thousands of dollars in compute on $200-a-month subscriptions, a move that drew a fair amount of backlash at the time. More recently, in March, Anthropic tightened five-hour session limits further during peak hours, citing surging demand that had pushed Claude’s daily active users up roughly 5x since the previous October. Anthropic was upfront that the change would cause around 7% of users, disproportionately on paid tiers, to hit caps they hadn’t run into before.
Against that backdrop, a full reset landing the same week as a competing frontier model launch is easy to read as a response, whether or not that’s actually what happened internally at Anthropic. Sottiaux’s read — that this looks like a company loosening the leash right when a rival just gave users a reason to look elsewhere — is the obvious one, and it’s clearly the one he wanted people to land on.
The Rivalry Has Been Getting Personal For A While
This isn’t the first time OpenAI and Anthropic staff have sparred publicly rather than letting the benchmarks do the talking. Sottiaux has form on rate limits specifically — back in May, when Codex crossed 5 million users, he announced OpenAI would reset usage limits the next morning to celebrate, framing it as a milestone moment rather than damage control. Seeing Anthropic run the same play this week, without a milestone attached to it, is presumably what made the timing look off to him. The GPT-5.6 release itself leaned into direct comparisons too, with OpenAI’s own launch materials placing Sol ahead of Claude Mythos on Terminal-Bench and several agentic benchmarks.
The numbers from independent evaluators this week actually back up some of the swagger. GPT-5.6 Sol landed just one point behind Claude Fable 5 on the Artificial Analysis Intelligence Index while costing roughly a third as much per task, and it topped the Coding Agent Index outright. Anthropic still holds the intelligence crown and leads on realistic knowledge-work benchmarks like AA-Briefcase, so there’s a real case underneath the trolling. Sottiaux’s tweet is doing what a lot of these exchanges do lately — turning a genuinely close competitive race into a moment built for screenshots, on a week where the underlying benchmark results already gave OpenAI something to point to.
Whether Anthropic’s rate limit reset actually had anything to do with GPT-5.6 landing, or was simply scheduled capacity work that happened to coincide with a bad week to look defensive, Anthropic hasn’t said. But the jibe did land, and as over 600 reposts would attest, found some resonance within the AI community on X.