AI Tools Finally Started Telling the Truth About Price
For two years every AI app hid its real cost behind 'from $X'. This month the math got honest.
GPT-6.1 Sol's pricing is stated plainly: $2 in, $10 out per million tokens, cached input at $0.10. That lets you actually model a bill before you ship. For a year, vendors led with the free tier and hoped you'd never read the API docs.
ElevenLabs' v4, out September 30, leads one arena and sits second on another, with a Turbo variant at roughly 100ms median latency across 90-plus languages. The honest part isn't the speed - it's that they published the arenas instead of a vague 'most emotive ever' claim.
Decision models take honesty furthest: per-call cost is fractions of a cent, stated up front, because there's no long output to bill. When the output is a single typed choice, the pricing can't hide.
I'll take stated pricing over a clever tagline any day. The tools that win my budget this quarter are the ones that let me predict the invoice.
Updated October 28, 2026: the stated prices below were verified at announcement.
Frequently asked questions:
Q: What did GPT-6.1 Sol's pricing look like?
A: Stated plainly: $2 in, $10 out per million tokens, cached input at $0.10 - so you can model a bill before you ship.
Q: What was honest about ElevenLabs v4?
A: They published the arena results instead of a vague 'most emotive ever' claim - leading one arena, second on another, with a Turbo variant near 100ms median latency across 90-plus languages.
Q: Which pricing is most transparent?
A: Decision models: per-call cost is fractions of a cent, stated up front, because there's no long output to bill.