The 24x Hidden Price Range

June 3, 2026 · 5 min read

You open Copilot. You ask it to review your code. It works. You move on.

What you don't see: Copilot's Auto Mode just picked one of several models to handle your request. The cheapest option costs 161 credits. The most expensive? 840 credits. That's a 5.2x difference for the same task — and Auto Mode doesn't tell you which one it chose.

But that's just the per-request spread. Across the full product, the hidden price range stretches to 24x. And in 12 days, Anthropic installs the same meter on your Claude credits.

The Price You Can't See

Ars Technica's deep-dive into Copilot's new billing model revealed something extraordinary: the same code review task costs between 161 and 840 credits depending on which model Copilot silently selects.

The Full 24x Range

The per-task variance (5.2x) is just the beginning. Across Copilot's entire pricing structure, the hidden range is much wider:

From 5 credits to 1,200+ credits. That's a 24x range. And Auto Mode decides where on that spectrum your request lands — without telling you.

The $6 Request

The Register reported that a single complex request can cost the equivalent of $6 in token charges. One request. Six dollars. The developer didn't choose the expensive model — Auto Mode did.

The Annual Plan They Took Away

Here's what makes this worse: you can't lock in a rate anymore.

GitHub confirmed in their official FAQ that annual plans are being retired. No more yearly pricing. No more predictability. Just month-to-month token billing with a 24x hidden range.

They also eliminated free models entirely: "Free models are no longer part of our offering." The safety net is gone.

Deepseek: The Contrast That Proves the Point

While Copilot charges up to 840 credits for a code review, Deepseek v4 Flash processes 15 million tokens for 7 cents. Same underlying technology. Radically different price.

The gap between what AI actually costs to run and what vendors charge is now a canyon. And it's about to get wider.

June 15: The Meter Comes to Claude

In 12 days, Anthropic splits its credit system. The credits you have today — the ones that cover Claude Code, Claude API, and all Anthropic products — will be reallocated with a new, more granular billing structure.

Sound familiar? It's the same playbook:

1. First: Simple, predictable pricing ("$20/month, all-you-can-use")
2. Then: Credits with a generous allocation
3. Then: Credits split into categories (coding vs. chat vs. agentic)
4. Then: Per-token billing with "Auto Mode" that picks the price
5. Finally: Annual plans retired, free tier eliminated, sign-ups paused

Anthropic is on step 3 right now. Copilot just completed step 5. The trajectory is identical.

The Alternative: Own Your AI

You don't have to rent AI by the token. You can own it.

• Flat-rate pricing — no meters, no credits, no "Auto Mode" surprise billing
• Model freedom — switch between GPT, Claude, Gemini, Deepseek, or local models without paying per switch
• No vendor lock-in — your tools, your data, your choice of model
• Countdown-proof — when June 15 arrives, your bill doesn't change

The 24x hidden price range isn't a bug. It's the business model. Every time Auto Mode picks a more expensive model, it picks your pocket. And there's no setting to turn it off.

Updated June 3, 2026 with Day 2 Copilot billing data and Anthropic IPO filing confirmation. Sources: Ars Technica, The Register, GitHub Community, Washington Post.

Related: The Landlord Is Selling the Building (Anthropic IPO) · Copilot Day One: Rent Arrived · The Safety Net They Removed · Anthropic June 15 Credit Split · Flat-Rate AI vs Token Billing