Anthropic doubles Claude Code cost
- Anthropic quietly raised Claude Code’s published daily token-cost estimate in late April, pushing the average active developer from $6 to $13. - The bigger jump was at the high end — Anthropic’s “90% of people” estimate moved from $12 to $30 per active day. - The change matters because Opus 4.7 is now the premium coding default, making AI coding spend less predictable for teams.
Anthropic just made one of the hidden costs of AI coding a lot more visible. Its Claude Code docs now peg the average token spend at about $13 per developer per active day, up from $6 earlier in April. The high-end estimate moved even faster — from $12 to $30 for 90% of users. That is not a small tweak. It is Anthropic telling customers that the economics of “just let the coding agent handle it” changed as newer, heavier models took over. (finance.yahoo.com) ### What exactly changed? The change was in Anthropic’s own public estimate for Claude Code usage, not in a splashy new pricing announcement. The company revised the expected token cost per active developer day upward sometime around mid-April 2026, and the new numbers (finance.yahoo.com) many teams use to budget before they have real internal usage data. (finance.yahoo.com) ### Why did the estimate jump so much? Basically, Claude Code is leaning harder on a more expensive model. Anthropic launched Claude Opus 4.7 on April 16, 2026 and positioned it as a stronger coding model for difficult engineering work. Opus 4.7 starts at $5 per million(finance.yahoo.com)5. If the product routes more coding workflows to Opus-class reasoning, daily spend rises fast. (anthropic.com) ### Why does “active day” matter? Because this is not a flat subscription story. Claude Code spend scales with how often a developer uses the agent and how long the workflow runs. An “active day” is really shorthand for days when the model is reading code, editing files, running commands, and often iterating across a repo. Claude Code(anthropic.com)s an agentic coding system that can make changes across files and run tests. More autonomy usually means more tokens. (anthropic.com) ### Why is the 90th-percentile number the scary one? Because budget pain usually comes from outliers, not averages. Moving the typical estimate from $6 to $13 is a little over 2x. Moving the “90% of people” estimate from $12 to $30 means heavier users could blow through internal budgets much faster than managers expected just weeks ago(anthropic.com) seats. One intense debugging session or long-running agent loop can create a very different bill from a quiet day. (finance.yahoo.com) ### Is Anthropic actually raising prices? Not in the simple sticker-price sense. The underlying model pricing Anthropic publishes is still explicit and per token. What changed here is the company’s estimate of what real developer behavior will cost inside Claude Code. T(finance.yahoo.com)a smarter but pricier model, the effective cost of using the product goes up even if the pricing table itself looks familiar. (anthropic.com) ### Why would teams still accept that? Because better coding agents can be worth it if they finish harder tasks with less supervision. Anthropic is selling Opus 4.7 as exactly that kind of upgrade — a model for complex software engineering and long-running agentic work. On its own benchmarks, Anthropic says Opus 4.7 resolves far more productio(anthropic.com)straightforward: higher token burn in exchange for fewer hand-holding loops from engineers. (anthropic.com) ### What does this mean for the AI coding market? It is a reminder that “AI replaces labor” often first shows up as “AI adds a volatile compute bill.” Coding agents are getting better, but they are also becoming more like cloud infrastructure — powerful, elastic, and easy to overspend on. Anthropic’s revised estimate makes that visible. The bo(anthropic.com)now the meter matters a lot more. (finance.yahoo.com)
