The Utility of Genius

The models arrived in a heap this week. Grok 4.5 landed as an Opus-class model at roughly sixty percent less than Opus. Anthropic shipped, OpenAI moved its next release toward the door, and a couple of open-weight labs matched last year’s frontier at roughly a hundredth of the price. The trade press called it a race heating up. I read it the other way. This is the sound a market makes when a thing stops being scarce.

The race is over, and nobody won. A race has a finish line and a leader who pulls away. What we have is a pack that keeps converging, each new front-runner beaten on price within a fortnight by whoever was just behind. Three frontier models dropped in a single day, which one observer called “not a race heating up, that’s the sound of models becoming a commodity.”

Muhammad Ali@get_Muham

When the power user can't tell models apart, the only differentiator left is price and distribution. We just watched AI models become commodities faster than anyone predicted.

Yuchen Jin @Yuchenj_UW

I’m the kind of man who can confidently pass a blind test between Coke, Pepsi, and Diet Coke. But if I’m brutally honest, I probably can’t pass a blind test between GPT-5.6 Sol, Claude Fable 5, and GLM-5.2 for 95% of my daily use cases.

6:08 AM · Jul 10, 2026

That is not a lead changing hands. That is a commodity forming.

Intelligence saturates, the way it always has. Here is the part the benchmark charts miss, and the reason none of this should surprise us. We ran this experiment already, with people. Average general intelligence is set at 100 by definition, and for the overwhelming majority of real work, somewhere past 120 or 140 the extra points stop paying. A brilliant lawyer and a merely good one file the same motion. Above the ceiling of the task, more IQ is a rounding error. Models are now crossing that same ceiling for most of what we ask them, and the people paying the bills have noticed. As one engineer wrote, “once the weekly model soap opera calms down, a lot of people will discover they never needed the frontier model anyway”

Shawn Chauhan@shawnchauhan1

Here's an uncomfortable stat for the "frontier or nothing" crowd. Open-weight models like GLM-5.2, DeepSeek-v4, and Kimi-K2.7 are now handling 70 to 80 percent of production AI workloads. Not because they're the smartest models on the leaderboard. Because routing the easy 80

5:52 AM · Jul 13, 2026 · 903 Views

5 Replies · 4 Reposts · 13 Likes

Opus versus Fable is a live question at the frontier and an immaterial one for the ninety percent of work that never goes there. AI, like intelligence itself, was always going to become a commodity. I have argued before that the model is necessary but not sufficient. This is what that looks like when it lands.

The gap is real, it is closing, and the best of it was never for sale. None of this means the frontier labs are in trouble at doing frontier things. The closed models still lead, by something like three to five hundred Elo points on the hardest work, and that lead is worth paying for when the work is hard enough.

Louis-François Bouchard 🎥🤖@Whats_AI

And here's how open models compare on the same benchmark (boxed = open weights, log cost axis): Kimi K2.6 Thinking is the open-weights headline: it edges past Grok 4.5 and beats GPT-5.5 Extra High at $0.05/task, ~8× cheaper. DeepSeek V4 Pro (Extra High) writes at Gemini 3.1 Pro

Louis-François Bouchard 🎥🤖 @Whats_AI

Early results from our internal benchmark for whether a model can write well and follow tone or not: Skip Sol Extra High entirely: it writes slightly worse than Sol default at 2.3× the price. Extra thinking helps small models write, not big ones (Luna +182 Elo, Terra +193,

11:08 PM · Jul 11, 2026 · 14K Views

11 Replies · 6 Reposts · 80 Likes

But the open models have caught up to where the closed frontier sat a year ago, at roughly one one-hundredth of the price, and a year is not much of a moat. If a lab ever built something genuinely world-changing, the last thing it would do is rent it to everyone by the token. It would keep it and use it. What reaches the API, by definition, is the tier a lab is willing to sell to all comers, which is another way of saying the commodity tier. The breakthroughs will keep coming, and they will still matter enormously. They just will not be the thing on the price list. Meanwhile, the price list gets longer: once a buyer has five vendors who all clear the bar, price falls, and so, in time, do the trillion-dollar valuations built on the premise that one of them owns something the others cannot copy. If the model is the commodity, it is worth asking out loud what exactly the trillion-dollar valuation is paying for. I flagged the price collapse coming when DeepSeek did it cheap early last year; the $1.25 trillion being staked on the idea that raw intelligence itself is the prize is the wager on the other side of this one.

DeanGuida@DeanGuida

Generic AI gives you generic answers. The leverage is the context layer: your data, your workflows, your IP. The model is the commodity. The context is the competitive advantage. #AI #AIStrategy #Leadership #EnterpriseAI

11:00 PM · Jul 13, 2026 · 5 Views

1 Like

The winner is not the one with the best model, or even the most intelligence. It is the human who fuses a cheap brilliant mind with a lifetime of tacit skill and a better memory of the problem, so that one plus one comes out to a good deal more than three.

I will admit I am living this, not just describing it. My own work now runs across a rotating mix of Claude, OpenAI, Grok, and GLM-5.2, in their cloud environments, driven from Cursor or the Claude command-line harness I have come to prefer. I am token-maxing: sending each task to whichever deal is cheapest for the quality it needs, and switching the moment the math changes. The models have become sufficiently interchangeable that the choice is now mostly economic. My productivity keeps climbing, the cost keeps falling, and my mind can barely keep pace with the tools. It is coping. Not one genius to rule them all, then, but a cheap, swappable bench of them, and a human in the chair learning, a little breathlessly, to conduct.

The old order priced the frontier model as the crown jewel and built the valuations to match. The new one treats it as current, cheap, and everywhere, and quietly moves the margin, and the meaning, to the layer above the pipe. The frontier is no longer a spot on a benchmark. It is the narrow, and very human, space between a rented mind and someone who knows what to ask it, and what to do once it answers. Those few are the same people ahead without the AI.