Faros AI’s guide to how much Claude Code costs does the plan arithmetic first: Pro at $20 a month, Max 20x at $200, Team seats from $20 per seat billed annually, Enterprise at $20 per seat plus usage at API rates, and Opus 5.5 at $4 per million input tokens and $20 per million output. Then it makes its real argument, which is that none of those numbers tell an engineering leader anything. Faros CEO Vitaly Gordon describes one Faros engineer spending $20,000 on tokens in a month, and “nobody could say whether that called for a promotion or a stern conversation, because the cost number carried no information about what value it produced.” The measure Faros wants instead is cost per outcome: spend tied to merged pull requests, resolved tickets and released features. Its slogan for the shift is “from token maxxing to outcome maxxing.”
The pricing tables are Anthropic’s list prices reproduced, and Anthropic’s own benchmark of $150 to $250 per developer per month and about $13 per active day is attributed to Anthropic. The vendor figure in the post is Faros’s: 211 of its own engineering tasks run through seven coding routes, with Claude Code on the open model GLM-5.2 scoring higher on a model judge than Claude Code on Opus 4.8 and Codex on GPT-5.5, at roughly half the projected token cost per task. That is a benchmark on one company’s repositories, scored by a model, in service of a product called Time Machine that automates the same test. What survives independent of the pitch is the advice beneath it: a cheaper per-token model can cost more per task, older Opus models cost more per token than Opus 5.5, and the only evidence that counts is a test on your own code.
The catalog shows what Faros ships to anyone who wants to run that test themselves. Faros AI on apis.io carries six API pages. The cost-per-outcome thesis depends on getting spend and delivery data into one model, and that is the Faros AI Ingestion API and the Faros AI Events API on the way in, and the Faros AI Graph QL API over the canonical model on the way out. The Faros AI Webhooks API is the push path. The catalog record corroborates the architecture the post implies: a REST surface for events and ingestion plus a GraphQL query API over a connected canonical model, with an open-source Community Edition. The agentic access profile counts 12 operations, six of them acting.
The measurement is where the story turns. Faros scores 20.2 on the Kin Score, emerging, and discoverability at 66.1 is doing almost all of the work. Contract quality is 9.0. Contract governance is 0.0. Operational transparency is 13.2 and access clarity is 15.8. Agent Readiness is 19.6, agent-aware, with spec presence, agentic access, auth clarity and rate limit signal lit, and error semantics, OpenAPI examples and MCP server all unlit. A company whose entire argument is that other people’s AI tooling needs visibility into what it produces publishes an API contract the rubric can barely see into. The gateway Faros would have you enforce routing policies through has no MCP server in the catalog and no documented error semantics, which is a hard place from which to tell anyone else their invoice carries no information.