← Back to feed

Anthropic announced on May 13 that all paid Claude subscribers will receive a dedicated monthly credit for programmatic usage starting June 15. The credit covers Agent SDK tasks, claude -p non-interactive CLI runs, and the official GitHub Actions integration. What it does not do is subsidise any of them.

Credit amounts match your subscription tier: $20/month for Claude Pro, $100/month for Claude Team, $200/month for Claude Max. All credits are drawn at API list-price rates, do not roll over at month-end, and cannot be topped up from your general interactive subscription pool. When the credit runs out, you buy additional usage credits or migrate to the API directly.

The move reverses — and reprices — the April 4 policy that had blocked third-party agents from using Claude subscriptions entirely.

Why Anthropic shut it down first

In early April, Anthropic prohibited programmatic usage from subscription tiers after discovering that autonomous agents running through third-party harnesses were consuming hundreds of dollars of token-equivalent capacity against subscriptions priced at $20–200/month. The root cause was cache efficiency. First-party tools like Claude Code and Cowork are engineered to maximise prompt cache hit rates, reusing previously processed context to save on expensive compute cycles. Third-party agents — including external harnesses like OpenClaw — typically bypass caching entirely, forcing full reprocessing on every request.

Boris Cherny, Head of Claude Code, acknowledged these external agents were “really hard for us to do sustainably.” Even with access to the 300MW Colossus 1 data centre and 220,000 GPUs, agentic demand was outpacing what flat-rate subscriptions could absorb.

The new pricing reality

The Agent SDK credit system solves the subsidy problem by eliminating the subsidy. Programmatic usage is now billed at full API list prices regardless of subscription tier. The monthly credit is an allowance at those prices, not a discount layered on top of subscription value.

For most users:

• Claude Pro ($20/month): $20 of Agent SDK credits at API rates. If programmatic usage was previously consuming more than $20/month of subsidised capacity — common for users running daily automation — the effective credit is now capped at actual plan value.
• Claude Team ($100/month): $100 credits. Covers moderate automated workflows.
• Claude Max ($200/month): $200 credits. Heavy users who were previously consuming disproportionate capacity get the most generous allowance, but it is still capped and billed at cost.

Independent analysis puts the effective price increase at 12x–175x for the median heavy programmatic user, depending on how cache-inefficient the workload is. A subscriber running a persistent agent that was consuming $500/month of token-equivalent capacity against a $20 subscription was operating at a 25x subsidy ratio. That ratio no longer exists.

What stays the same

Interactive usage — Claude Code in the terminal or IDE, Claude.com, and Cowork — remains on existing subscription limits, which Anthropic raised 50% through July 13. The two budgets are independent. Interactive overages do not draw from Agent SDK credits, and Agent SDK exhaustion does not affect interactive limits.

For developers running lightweight automated tasks — daily summaries, a low-frequency GitHub Actions bot, occasional claude -p scripts — the $20–100 monthly credit likely covers the workload at list prices. For anyone running persistent agents, high-frequency programmatic pipelines, or workflows that were previously viable only because the subscription subsidised token costs, the API is now the correct billing tier.

The June 15 effective date gives developers five weeks to audit workloads, estimate actual API costs, and decide whether to restructure, reduce volume, or migrate to direct API billing.