The Rankings Just Got Weird — 3 Surprises From This Week's Data
I’ve been watching these rankings long enough that most weeks are boring. A few spots up, a few down, nothing to write about.
This week wasn’t boring.
Surprise #1: A model nobody saw coming
There’s a model I won’t name yet — I want another week of data before I call it a trend. But it jumped 12 spots this week. Twelve. That doesn’t happen by accident.
What I can tell you: it’s a mid-sized open-source model, it’s priced aggressively, and the growth is coming entirely from code-related usage. Developers found something and they’re not telling anyone yet.
I’ll name it next week if the numbers hold.
Surprise #2: OpenRouter pricing shifted overnight
Three models changed their per-token pricing on OpenRouter this week. All three went down. Not by 5% or 10% — one dropped 30%.
The platform competition is real. SiliconFlow, Groq, and now OpenRouter are all adjusting in response to each other. If you hardcoded a model and price into your app three months ago, check the current numbers. You might be overpaying by a lot.
This is why I built the cross-platform price table on every model page. Prices move. The table keeps up.
Surprise #3: The #1 spot isn’t as safe as it looks
Claude Opus 4.6 held #1 again. But the margin shrunk. Not by a little — usage share dropped from 18.2% to 16.7% of the top 50. That’s the biggest single-week drop I’ve recorded.
Where did the users go? Split across three open-source models, all in the top 20. Nobody is catching Claude tomorrow. But the trend line says someone might next quarter.
What changed in the top 20
- ↑ Mistral Large 3 — #14 → #11. Code benchmarks are driving this. If you’re building developer tools, pay attention.
- ↑ DeepSeek V4 — held top 10, usage up 8%. Steady, not flashy.
- ↓ GPT-4.1 — dropped 3 spots. Not a quality issue — pricing. Users are finding cheaper alternatives for the same tasks.
- ↑ Qwen 3 — cracked the top 15 for the first time. Strong vision capabilities are the draw.
This month’s theme: the price-quality line is blurring
Six months ago, you paid a premium for quality. The expensive models were noticeably better.
That’s changing. The gap between a $2/million token model and a $0.40/million token model is smaller than it’s ever been. For a lot of tasks — not all, but a lot — you can’t tell the difference anymore.
If you’re still using defaults from 2025, this is your sign to re-evaluate.
What I’m watching in June
Three things:
- Can anyone dethrone Claude? Not this month. But the trend is real.
- The anonymous 12-spot jumper. One more week of data and I’ll tell you who it is.
- Price war acceleration. At this rate, the average price per token could drop another 15-20% by July.
Check back next Monday. If the data holds, this summer is going to be interesting.
Data from OpenRouter rankings, week ending June 1, 2026. Cross-referenced with SiliconFlow and Groq pricing where available.