← ← Back to Blog

The AI Price War Is Real — Here's the Receipts

The AI Price War Is Real — Here's the Receipts
rankingsweeklyanalysispricing

April was brutal for AI pricing. In a good way, if you’re a buyer.

Average price per million tokens across the top 50: down 11% month over month. That’s on top of a 22% drop in Q1. The trend line is unmistakable.

Who’s cutting and why

SiliconFlow started the latest round. They cut prices on 8 models in April alone, some by as much as 40%. The strategy is transparent: buy market share now, monetize later.

Groq matched within days. Groq’s advantage is hardware — their custom chips give them unit economics that software-only platforms can’t touch. They can cut and still make money.

OpenRouter didn’t cut directly but added three new providers to their routing layer, which pushed effective prices down about 7% through competition alone.

The models getting cheapest fastest

Entry-level models are approaching zero:

  • Small open-source models: some now under $0.05/million tokens
  • Mid-tier models (Mistral class): $0.15-$0.30 range, down from $0.30-$0.50
  • Premium tier (Claude, GPT-4): holding firmer, but still down 5-8%

The biggest percentage drops are in the mid-tier. That’s where the competitive pressure is highest.

What this means for your bill

If you locked in a model and a provider six months ago, you’re probably overpaying by 30-40%. Not because anyone raised prices on you — because everyone else lowered theirs and you didn’t notice.

Go to the model page. Check the cross-platform price table. Spend 5 minutes. Save actual money.

How low can this go

There’s a floor. Inference costs real compute, and GPUs aren’t free. But we haven’t hit the floor yet. My estimate: another 25-35% decline across the top 50 by end of year before things stabilize.

The beneficiaries are builders. Every dollar you save on inference is a dollar you can spend on better features, more experiments, or — imagine — profit.

This week’s rankings note

Quiet week for rank changes. The big mover was a small open-source coding model that jumped 7 spots. I’m tracking it but want another week before writing it up.

The price chart is on the Compare page if you want to see this visually.