Claude Code Pricing: Sonnet 5, Plan Access, and API Costs

Buying path

Claude Code pricing starts with the access route. Some developers use Claude Code through eligible Claude plan access. Other teams build or automate around Anthropic API and Console billing. Those are different budget lanes, and the Sonnet 5 and Fable 5 launches affect the Claude-side model-budget conversation.

The practical first question is where the work happens. If a developer is interactively using Claude Code on a repo, plan access may be the right starting point. If a product, internal platform, or automated agent calls Claude models directly, API token pricing is the budget surface.

Model budget after Sonnet 5 and Fable 5

For API-priced work, Sonnet 5 should be treated as the current lower-cost agentic and coding route. Anthropic lists introductory Sonnet 5 pricing at $2 per million input tokens and $10 per million output tokens through August 31, 2026, then $3 and $15. Fable 5 stays the higher-cost hard-task escalation lane at $10 and $50.

Do not turn that into a Claude Code default claim. Unless official Claude Code release notes or account behavior make the model route explicit for a buyer's access path, write the budget as Claude Code access plus possible API/model-routing costs.

The useful eval is not whether Fable sounds smarter. Measure whether the route produces accepted patches, reduces retries, handles larger context, or saves engineering time on the hardest jobs. If it does, the higher price can be justified. If not, Sonnet 5 or Opus 4.8 may remain the better Claude baseline.

Small code explanations, one-file edits, routine refactors, and low-risk helper tasks often do not need the most expensive model lane. Keep Sonnet 5 or another stable route for normal work and reserve Fable testing for tasks that repeatedly stretch the current setup.

Team controls and Mythos access

Team pricing should include governance, not just tokens or seats. Admins need to decide who can run long agentic jobs, which repositories are in scope, how output is reviewed, and how spend is monitored. Review time is part of the cost because generated code still has to survive tests, code review, and deployment rules.

Mythos 5 is a separate question. It should not be treated as a normal Claude Code pricing option for ordinary teams. If an organization has a reason to evaluate restricted Mythos-class access, that belongs in an enterprise, security, and procurement conversation before any budget assumes availability.

Upgrade boundary

Buy Claude Code access when serious repo work is repeated enough to justify an agentic workflow. Add API budgeting when software or automation needs programmable Claude calls. Test Sonnet 5 first when the actual access path exposes it, and reserve Fable 5 for task difficulty that justifies that model lane.

Do not let model names replace budgeting discipline. The safest rule is to separate plan access, API spend, model routing, and review effort. Claude Code is easiest to justify when it turns hard coding work into accepted, reviewed changes with less total engineering overhead.