Anthropic Cut Its Flat Subscription Price and Killed the API Discount. Sales Orgs Buying Claude Now Need a FinOps Plan.

Anthropic replaced flat-seat subscriptions with consumption-based billing in May 2026, and Forrester analyst Tracy Woo’s read is that the change moves AI cost risk back onto the customer. The old model charged $200 a month for premium users and $40 for standard, with API discounts of 10-15% built in. The new model drops the sticker price — $10 a month for Claude.ai, $20 for Claude Code — but removes the API discount entirely, and Agent SDK credits expire monthly with no rollover: once a team burns through its credits, usage either stops or falls back to standard API pricing.

Woo’s point, quoted directly, is that “token usage, model selection, Claude Code adoption, and commitment levels will have a far greater impact on spend than seat counts.” That inverts the old budgeting math. A lower per-seat price looks like savings on a procurement spreadsheet, but a small cluster of heavy users — the SDR running Claude Code against every account in a territory, or an AI-SDR workflow calling the API on every enriched lead — can now drive spend that has nothing to do with headcount. Forrester’s piece notes OpenAI, Google, and Cohere are moving toward similar consumption-based structures, so this isn’t an Anthropic-specific quirk sales and RevOps buyers can route around by picking a different vendor.

Forrester has already published the playbook for diagnosing exactly this kind of overrun. In a companion Forrester piece on token-spend variance, analyst Greg Zorella argues that “too many tokens” isn’t a real diagnosis and proposes splitting spend variance into rate (price per token), volume (tokens consumed), and mix (which model tiers and token types are being used) — with a concrete first step of confirming usage is actually trackable by token type before building a dashboard on top of it.

For any sales org buying Claude seats to run AI-assisted outbound, the practical move is to stop treating the monthly subscription line as the budget and start tracking token consumption by workflow before a finance review forces the question.