Claude Pricing: Plans, Sonnet 5, Fable 5, and API Costs

Buying path

Start Claude pricing by separating app access from API usage. Claude app buyers compare Free, Pro, Max, Team, and Enterprise-style subscription routes. Developers and product teams compare Anthropic API token pricing. Those lanes can support the same broader Claude strategy, but they do not share one simple price table.

That split matters more after Sonnet 5, Fable 5, and Mythos 5. A Claude app subscription may be the right purchase for a person doing writing, research, documents, or coding help in Claude. An API budget is the right purchase for a product, internal agent, or automated workflow that calls specific Claude models directly.

App subscriptions and team access

For most individuals, Claude Pro remains the normal paid entry point. Max is for heavier personal use, and Team or Enterprise routes make sense when administration, workspace governance, procurement, or policy controls matter. This part of the decision should be judged by user seats, message limits, workspace controls, and whether Claude is actually used every day.

Claude Code can also sit near the app subscription conversation when a developer uses eligible Claude access interactively. That does not erase API pricing. Teams should ask whether the work happens in the Claude product experience, in Claude Code, through Console/API keys, or across several lanes.

API pricing, Sonnet 5, and Fable 5

For API buyers, the important update is model routing. Anthropic's current pricing lists Sonnet 5 at an introductory $2 per million input tokens and $10 per million output tokens through August 31, 2026, then $3 and $15 after that. Opus 4.8 remains listed at $5 and $25, while Fable 5 is the higher-cost escalation lane at $10 and $50. That makes Sonnet 5 the first API model many agentic and coding teams should test before moving harder work to Fable 5.

The right API budget test is cost per accepted result. If Fable 5 solves a difficult task in fewer retries, handles longer context better, or produces a more usable final output, the higher token price may be justified. If the workload is classification, extraction, short chat, or routine drafting, Sonnet 5, Opus 4.8, or cheaper Claude models may be the better route.

Mythos 5 should be handled differently. It is restricted-access, so it does not belong in ordinary self-serve budget planning. Enterprise and security teams should treat Mythos as an approval path that may require access review, data-handling review, and monitoring expectations.

Upgrade boundary

Choose a Claude app plan when humans need a better daily assistant. Choose API billing when software needs programmable model calls. Test Sonnet 5 first for agentic or coding work where price matters, and choose Fable 5 only for the slice of API or agent work where quality, context, or output length can justify higher spend.

Do not compare Claude pricing by headline numbers alone. Compare the buying route, the owner of the budget, the model used, and the review cost around the output. That is the difference between a clean Claude rollout and a bill that feels surprising two weeks later.