Cognitive Warmup. Some ideas to get you started. Random, unrelated thoughts, but I wanted to get them off my chest.
It was in August 2025 when AI startup Perplexity bid $34.5 billion to buy the Chrome browser from Google. Arvind Srinivas was all over social media feeds, and you wouldn’t have been criticized for feeling that he’s larger than life (and that perplexity is amazing). Here we are, six months later, and the latest we’ve heard from Perplexity is the Model Council.
How times change, when the bubble has to be kept inflated.
Secondly, it is being reported that Nvidia Corp will stop new gaming GPU releases in 2026. Reason? The reason for this is believed to be the global memory chip shortage. Don’t you say that, Nvidia? The same shortage (and price gouging, mind you) that you played a big role in creating?
algorithm
This week, on Neural Dispatch:
- Discussion on Perplexity’s AI Model Council
- Understanding Anthropic’s Opus 4.6 Model
- Meta Platform’s Avocado Diet.
A council of AI models deliberating on your query
Perplexity’s Model Council, what the AI company dubs a multi-model research facility that brings together multiple models for an answer, is perhaps a push for that subjective relevance and accuracy to specific questions that should have been tackled long ago.
Call me an optimistic pessimist by nature, but having large language models (LLMs) with multiple frontier logic at work, does not increase the chances of getting accurate (and correct) information – different AI models have different endpoints, but what you are likely to get is a variety of approaches, calculations or deductions. I wouldn’t say that anything about AI is 100% correct, because that would imply that AI is absolutely perfect (which is not the case, almost always).
Perplexity says that the three models will work on a prompt or query if the user selects Model Council (for example, the query will run in parallel on Cloud Opus 4.6, GPT 5.2, and Gemini 3.0), relying on a synthesizer model that reviews the output, resolves conflicts where possible, and returns an answer to the user that shows where the models agree, and where they differ. I would like you to consider a few things.
- Each AI model, as I have already mentioned, has strengths and weaknesses. By having more than one model working on a query, the chances of ending up with gaps are greatly reduced.
- I would say that cross-validating the findings, creative brainstorming such as preparing travel plans and researching information would be areas where the Model Council methodology would work well.
- Model Council is a better way to work on multiple models, rather than having to switch between them.
- For now, don’t assume you’ll get access to the Perplexity Model Council without paying big bucks for it — it’s exclusive to the Max subscription tier, which will cost $200 (approximately) per month. ₹18,200).
Is Anthropic’s Opus 4.6 model better than anything else?
Anthropic PBC has released Opus 4.6, which it calls its “most intelligent model” and a “direct upgrade from Opus 4.5.”
This AI model is clearly well suited for enterprise and knowledge work, and improves on its predecessor in deep reasoning as well as planning, a large context window of up to 1 million tokens (the first Opus-class model to do so, which is good news for working with large datasets and documents), strong coding skills, and what Anthropic emphasizes are adaptive thinking skills.
Of course, as is the case with any model release, there are always multiple benchmarks neatly plotted in the graph to show superiority. Opus 4.6 is rated better than anything else, including Gemini 3 Pro and ChatGPT 5.2 in GDPval-AA (for knowledge task), BrowseComp (for agentive search), Terminal-Bench 2.0 (for coding) and Humanities Last Exam test (for multidisciplinary reasoning).
Anthropic says, “Opus 4.6 often thinks more deeply and revisits its reasoning more carefully before deciding on an answer.” Sure.
Is the avocado ripe yet?
Meta Platform Inc. is looking into AI rivals Google, Anthropic and OpenAI are likely to feel pressure, particularly from Google which has extended its lead from 2025 to this year, and Anthropic in particular has dialed up the enterprise focus.
Meta is believed to have distributed a new model internally, a proprietary model, called Avocado. Keep in mind, just last year CEO Mark Zuckerberg and Meta AI communications pitch called the Llama model “the most advanced in the industry.” Additionally, in the past year, Meta has launched a multi-billion dollar hire to fast-track its AI efforts.
Hopefully Avocado will release in some form later this summer, and we’ll get clarity on this model as time goes on. Do you know where to stay? Here, at Neural Dispatch.
Thinking
There is a reference to this. In 2024, OpenAI CEO Sam Attman rejected the idea of feeding chatbots with ads. Then he called it “the last resort”. Fast forward to now, ChatGPT will find itself hosting advertisements for free and Go tier subscribers, as it has become clear that the AI brothers have completely misunderstood how much users would be willing to pay for an AI subscription each month.
This is a case of growth of well-paid customers slowing down, and even if that’s not the case, this revenue stream can’t exactly fuel the infrastructure spending palace 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 the cloud.” He didn’t mention OpenAI by name, but it’s never a good sign when OpenAI immediately went defensive as a result. And also when someone insists they laugh, and are not angry at all.
A reality check: The moment Altman says, “Anthropic serves rich people an expensive product,” he is making it clear to everyone that AI companies will inevitably find themselves cornered: Plan A charges users a significant monthly fee and, therefore, convinces investors that everything is fine, and so, it is now onto Plan B where revenue from advertising tops the circular economy (I reported back in November that the AI math is broken and hopefully That bubble will remain inflated as long as possible.
Of course, we don’t even know how users will react once the ads go live in ChatGPTT. And then Altman goes on to say that Anthropic blocks companies they don’t like from using their coding product. Dirty linen being washed publicly is not a good way to convince investors that all is well in the sector. Nor does it do much to keep the bubble inflated. Well, recently, many AI “leaders” have begun to get quite angry, and very quickly. 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 provides curated insights on the critical technologies, practical applications and strategic implications shaping our digital future.
Want this newsletter sent to your inbox? Subscribe here.
