Neural Dispatch: Perplexity’s Model Council, Meta’s Avocado and Super Bowl ads

Cognitive warmup. A few thoughts to start off with. Random, unrelated musings, but wanted to get these off my chest.

(Unsplash)

It was in August of 2025 when AI startup Perplexity made a cheeky $34.5-billion bid to buy the Chrome browser from Google. Aravind Srinivas was all over social media feeds, and you wouldn’t have been criticised for feeling he’s larger than life (and that Perplexity is awesome). Here we are, six months later, and the latest peep we have heard from Perplexity is the Model Council.

How times change, when a bubble is to be kept inflated.

Secondly, it is being reported that Nvidia Corp. will pause new gaming GPU releases in 2026. The reason? A global memory chip shortage is believed to be the reason. You don’t say, Nvidia? The same shortage (and driving up prices, mind you) that you played a big role in causing?

Call me an optimistic pessimist by nature, but to have a number of frontier reasoning large language models (LLMs) at work, the chances of getting precise (and correct) information do not increase—different AI models have different gaps, but what you’ll likely get are a variety of viewpoints, calculations or deductions. I wouldn’t say 100% regarding anything about AI is spot on, because that would mean assuming AI is absolutely right (which it isn’t, almost ever).

Perplexity says three models will work on a prompt or query if the user selects the Model Council (for instance, the query would run across Claude Opus 4.6, GPT 5.2, and Gemini 3.0 in parallel), which is reliant on a synthesiser model that reviews the outputs, resolves conflicts where possible, and gives the user an answer that shows where the models agree, and where they differ. I’d like you to ponder over a few things.

This AI model is apparently suited for enterprise and knowledge work, and improves on its predecessor in deep reasoning as well as planning, a bigger context window of up to 1 million tokens (the first Opus-class model to do so) which is good news for working with larger datasets and documents), strong coding skills and what Anthropic insists is adaptive thinking skills.

Of course, as is the case with any model release, there are always a number of benchmarks neatly depicted in graphs to show superiority. Opus 4.6 is supposedly better than anything else including Gemini 3 Pro and ChatGPT 5.2 in GDPval-AA (for knowledge work), BrowseComp (for agentic search), Terminal-Bench 2.0 (for coding) and Humanity’s Last Exam tests (for multi-disciplinary reasoning).

“Opus 4.6 often thinks more deeply and more carefully revisits its reasoning before settling on an answer,” Anthropic says. Sure.

Meta is believed to have internally delivered a new model, a proprietary one at that, dubbed Avocado. Mind you, it was only last year that CEO Mark Zuckerberg and the Meta AI communications pitch had called the Llama models “most advanced in the industry”. Also, in the last year, Meta has embarked on a multi-billion-dollar hiring spree to fast-track its AI efforts.

Expect Avocado to release in some form later this summer, and we’ll increasingly get clarity on this model as time goes by. You know where to stay tuned? Here, on Neural Dispatch.

THINKING

There’s context to this. In 2024, OpenAI CEO Sam Atman had resoundingly dismissed the idea of stuffing advertisements into the chatbot. He had then called it a “last resort” then. Fast forward to now, ChatGPT will find itself hosting ads for free and Go tier subscribers, as it becomes clear the AI bros grossly misunderstood how much users will be willing to pay for AI subscriptions every month.

This is well a case of paid subscriber growth slowing down, and even if it isn’t, this revenue stream can absolutely not hold up the castle of infrastructure expenditure that OpenAI is trying to build.

Then Dario Amodei-led Anthropic released Super Bowl ads with the tagline—“Ads are coming to AI. But not to Claude.” They didn’t name OpenAI, but it is never a good sign when OpenAI immediately went into the defensive as a result. And also when someone insists they’re laughing, and aren’t annoyed at all.

A reality check: The moment Altman says “Anthropic serves an expensive product to rich people”, he’s making it clear for everyone to see that AI companies essentially find themselves cornered: The Plan A to charge a significant monthly fee from users and, therefore, convince investors that all is well hasn’t exactly worked out, and, therefore, it is now on to Plan B where revenue from advertising tops up the circular economy (I’d pointed this out as far back as November, that the AI math is broken and hope the bubble remains inflated for as long as possible.

To be sure, we don’t even know how users will react, once the ads go live within ChatGPT. And then Altman goes on to say that Anthropic blocks companies they don’t like from using their coding product. Dirty linen, being washed in public, is never a good route to convince investors all is well in the space. And neither does it do much to keep the bubble inflated. By the way, off late, many an AI “leader” seems to be getting quite annoyed, and rather quickly too. I wonder why that is.

Edited and produced by Tushar Deep Singh.

Neural Dispatch is your weekly guide to the rapidly evolving landscape of artificial intelligence. Each edition delivers curated insights on breakthrough technologies, practical applications, and strategic implications shaping our digital future.

Want this newsletter delivered in your inbox? Subscribe here.