What Is The Best AI Model In 2026? | The Brainstorm 140

**SPEAKER_1** (0:00)
Welcome to The Brainstorm. We've got Fable 5, we've got Chat GPT 5.6, Soul, Terra, Luna. We have Alex Carp on CNBC going off, like eight different people sent me that clip, and open source models coming in to scoop up some of the economics here.
Brett, maybe we go to you first with what was announced last week, and how are you comparing it?

**Brett Winton** (0:29)
Sure. So GPT 5.6 is kind of one, it's half the cost for the same performance of GPT 5.5, and two, it offers, I mean, at least also half the cost for the same performance as Fable, the Anthropics reputed best-in-class model, and offers higher-end performance at least on many benchmarks. So it's a clear catch-up and surpassing by ChatGPT, and I think that given people are focused on economics now, it'll probably continue to shift the marginal user over to open AI. But Frank, what do you think?

**Frank Downing** (1:11)
Yeah, I agree. I think, well, there have been a lot of model releases, like all the major, like five top US AI companies have all released models in the past month and a half, basically, from like Google's Gemini 3.5 flash launch through Meta releasing a new model that's actually better than that and a lower cost, which is really interesting. Grok 4.5, which is behind, but not that far behind, where Anthropic and Open AI are, and at a much lower cost. So you can clearly see the emphasis on efficiency that's going on right now, and the comparison relative to Anthropic, which it's still unclear to me, and they keep it unclear how much of the higher cost is them charging a premium because they can, versus their model is actually less efficient. We actually don't know the underlying efficiency. They might just be taking a much higher margin, but I would suspect it's a little bit of both.
The other thing is that the products are evolving rapidly as well. So Anthropic launched Claude Cowork, which has been their kind of lead out enterprise product for web and mobile, which makes it much more useful than having to sit next to your open laptop. OpenAid did the same thing by launching ChatGPT Work, which basically brings a codex-powered agent into all the ChatGPT surfaces.
And XAI is also iterating on the GrokBuild app, which is their coding agent. Also not as widely discussed because we're talking about all these frontier models, but the new ChatGPT voice capability is really good. I don't know if you guys have used it, but they basically came up with a way to let… Well, one, the voice model is much better. It's full duplex, so it can talk and listen at the same time. It can also delegate its thinking and tool use to a frontier model, like language model behind the scenes, which means you don't get these responses that are just totally dumb or fabricated. It sounds and acts like a frontier intelligence now that I think is just going to be like extremely useful.

**SPEAKER_1** (3:14)
Frank, why don't you talk us through this chart that you made comparing all of these performance metrics and cost?

**Frank Downing** (3:21)
Yeah, well, it occurred to me, I was kind of wondering, the genesis of this analysis was wondering when Gemini 3.5 Pro was going to come out. That's really the model that Google will expect to compete with Fable and GPT-5 6 Soul. And they did the same thing last year. They released Gemini 3 Flash. And then almost 30 days later, they released the Pro version. I think they were planning to do the same now. The Pro version of 3.5 has apparently been delayed. It's almost two months now since 3.5 Flash. I think no doubt they're trying to refine it based on the launches that have happened since then, which again, are from all of their biggest competitors, and they're increasingly competitive in terms of performance and cost.
And looking at the chart, you can kind of see what I walked through earlier, where OpenAI and Anthropic are clearly still in the lead in terms of frontier intelligence. As Brett said, OpenAI's latest model appears to deliver similar intelligence to Fable, but at less than half the cost. And then XAI, with the help of the cursor, both team and data set, training on XAI's infrastructure has delivered a near frontier intelligence at a much lower cost as well.
And Meta's back on the leaderboard and actually ahead of Google, just based on these benchmarks, which I think we can all agree are relatively one-dimensional and don't represent the real world utility of these models in as compelling of a way as you'd want them to. But it is the best basis of comparison that we have outside of vibes.