We maintain a pricing breakdown for Aider — the free, open-source terminal AI pair programmer where your real cost is whatever you pay the LLM API behind it. Checking it today against current model pricing before syndicating it, we found the entire cost calculator is built on a model that Anthropic replaced days ago, at a price Anthropic no longer charges.
What the page says
Our Aider pricing guide recommends Claude 3.7 Sonnet as the best-performing model for coding, at "~$3.00 per million input tokens, ~$15.00 per million output tokens." Every worked example on the page — daily spend ranges, monthly cost tiers, the head-to-head against Cursor Pro's flat $20/mo, the per-message cost of a large refactor — is built on that $3/$15 rate. The FAQ still states flatly: "Claude 3.7 Sonnet is currently the best overall performer for coding tasks in Aider."
What's actually true
Anthropic shipped Claude Sonnet 5.5 on 28 September 2026 as the new default Sonnet model, with a 1M-token context window — and it's priced at $2 per million input tokens / $10 per million output tokens, down from the ~$3/$15 the page still quotes. That's roughly a third cheaper, not a rounding difference. "Claude 3.7 Sonnet" isn't a stale price on a current model; it's two full model generations behind what a developer typing aider --model claude gets today.
This isn't a guess — it's a confirmed finding from our own internal freshness tracking, sourced to Anthropic's own Claude Code release notes (v2.1.284) announcing the Sonnet 5.5 launch and its pricing.
Why it matters for the page's own numbers
Every dollar figure on this page that assumes "Claude 3.7 Sonnet at $3/$15" is currently overstating real-world cost by roughly a third for anyone who configures Aider with today's actual default Sonnet model:
- The "moderate use" estimate of $30-$70/month becomes closer to $20-$47/month.
- The "$0.15-$0.30 per message" input-context cost for a 10-file project becomes $0.10-$0.20.
- The Aider-vs-Cursor-Pro breakeven math — the page's central argument for when Aider is cheaper than a $20/mo subscription — shifts in Aider's favor, not against it, once you use current pricing.
The direction of the error is consistent: the page has been making Aider look more expensive than it currently is, for months, because it never revisited the one number the whole page is built on.
The honest caveat
We haven't re-verified every other model row on the page (GPT-4o, GPT-4o-mini, DeepSeek) against current vendor pricing — this correction is specific to the Claude figures, which is what our tracking actually confirmed today. We're not going to quietly swap in a new number and pretend the old one wasn't there; the source page will get a full pass, not a patch, and until then treat the Claude-Sonnet figures on the live page as the outdated ones described here.
The actual lesson
Aider's pricing model — bring your own API key, pay per token — means our own cost breakdown ages the moment the vendor repriced. A flat-rate tool like Cursor or Copilot can go stale on features; a BYOK tool's pricing page goes stale on someone else's rate card, on someone else's schedule, and nothing on our site pings us when that happens except actually rechecking.
Full page (correction pending): devtoolsreview.com/pricing/aider-pricing
Top comments (0)