Claude Sonnet 4 + Opus 4 Retire June 15 — Migrate Before Requests Fail

Two breaking changes land on the same day — June 15, 2026 — and both are silent failures if you do nothing. Operators running anything built on Claude have four days to act. This article covers exactly what changes, what breaks, and what to do before the deadline.

The model retirement: two IDs go dark on June 15. Anthropic is retiring claude-sonnet-4-20250514 and claude-opus-4-20250514 on June 15, 2026. These model IDs were deprecated on April 14, 2026; June 15 is the hard retirement date. The official replacements are claude-sonnet-4-6 (replacing claude-sonnet-4-20250514) and claude-opus-4-8 (replacing claude-opus-4-20250514). Anthropic's documentation states directly: "Requests to models past the retirement date will fail." There is no grace period, no automatic redirect, and no fallback to a newer model. An API call to a retired ID does not silently route to the replacement — it fails.

What doing nothing looks like in production. Any application, pipeline, or config file that still references either of the two retired IDs will begin returning errors on June 15. This includes hardcoded model strings in application code, model IDs stored in environment variables or config files, defaults set in framework or library wrappers, and any third-party integration that passes a model ID on your behalf. The failure is silent until it happens — there is no deprecation warning at request time — so teams that haven't audited their codebase since April may not know they're exposed.

The Agent SDK billing change: a new credit pool, same day. Separately, effective June 15, 2026, usage from the Claude Agent SDK and the headless claude -p command — and CI jobs and automation built on the Agent SDK — will draw from a new monthly "Agent SDK credit" pool for subscription users. This pool is separate from interactive subscription usage limits. Monthly credit amounts by tier: Pro $20, Max 5x $100, Max 20x $200 (Team Standard seats $20, Team Premium $100; some Enterprise Standard seats are excluded). Credits are per-user — they cannot be shared or pooled across a team. They do not roll over: the balance resets each billing cycle. Once the monthly credit is depleted, additional Agent SDK usage flows to usage credits at standard API list rates, or stops if usage credits are not enabled.

What is NOT affected. Interactive AI coding tools use in the terminal or IDE — Claude Code running interactively (see our Cursor vs Claude Code comparison) — is not subject to the new credit pool. Neither is claude.ai web, desktop, or mobile chat, nor Claude Cowork. API-key users (direct API billing, not subscription plans) are also unaffected and continue under existing API billing. The credit pool applies specifically to subscription-plan users whose headless/automated usage runs through the Agent SDK or claude -p.

The operator impact. Before June 15, the retirement and billing change together represent two distinct exposure categories for most teams. A team with the old model IDs hardcoded in production will see broken API calls starting June 15 regardless of their billing setup. A team running heavy claude -p or Agent SDK automation on a Pro or Max subscription will hit their credit ceiling faster than they expect — particularly if they've been treating headless usage as covered by the same pool as interactive use. Both risks are fixable now with a targeted audit; both become production incidents if left unaddressed.