Two things are changing in the Anthropic ecosystem on June 15, 2026. If your team builds with Claude — whether through the API, Claude Code, or third-party tools that use the Agent SDK — both of them could affect you. Here is what you need to know and what to do before the deadline.

The Model Retirements

Anthropic announced on April 14, 2026 that two original Claude 4 models will be retired from the Claude API on June 15:

• claude-sonnet-4-20250514 — retire to claude-sonnet-4-6
• claude-opus-4-20250514 — retire to claude-opus-4-8

After June 15, any API calls to these models will return an error. There is no grace period and no gradual wind-down. The switch flips and requests fail.

This is standard practice for Anthropic — they give 60 days notice, then close the door. If you missed the April 14 announcement, you have roughly two weeks to act.

Who Is Affected

If you hardcode model IDs in your code and you are using claude-sonnet-4-20250514 or claude-opus-4-20250514, you will hit errors on June 16. That includes production applications, scheduled automations, internal tools, and any API integrations your team built over the past year.

The fastest way to check: export your usage data from the Claude Console (Settings > Usage > Export), review the CSV for any calls still hitting the deprecated model IDs, and track down the code that is sending them.

What You Are Upgrading To

Sonnet 4 to Sonnet 4.6

Claude Sonnet 4.6 is the current recommended balanced model for everyday tasks and agentic search. It consumes fewer tokens than Sonnet 4 while improving performance on multi-step tasks. Prompt caching is fully supported at standard pricing.

Opus 4 to Opus 4.8

This is a meaningful jump. Claude Opus 4.8 — launched May 28 — is Anthropic’s current flagship model. Compared to the original Opus 4:

• 1M token context window by default (compared to 200k on Opus 4)
• 128k max output tokens
• Adaptive thinking: the model decides per-turn whether to reason, reducing wasted thinking tokens
• Better long-horizon agentic coding, fewer derailments on complex tasks
• Mid-conversation system messages, useful for long-running agentic sessions

The catch is that Opus 4.8 dropped support for manually setting temperature, top_p, and top_k — a constraint inherited from Opus 4.7. If you set those parameters explicitly, you will get a 400 error. Replace them with prompting to guide model behavior instead.

For most teams, dropping in the new model ID and removing explicit sampling parameters is enough to get moving. The Anthropic migration guide has the full checklist.

The Agent SDK Billing Split

Starting the same day — June 15 — Anthropic is separating programmatic Claude usage from interactive subscription limits.

Until now, if you used Claude Code, the Agent SDK, GitHub Actions, or third-party tools like OpenClaw through a Claude subscription, all of that usage drew from your subscription’s monthly limits. That is changing.

From June 15, programmatic usage gets its own monthly credit pool, billed at standard API rates:

Plan	Monthly programmatic credit
Pro	$20
Max 5x	$100
Max 20x	$200

Interactive usage — chatting with Claude on the web, desktop, or mobile, and using Claude Code in the terminal — still runs against your regular subscription limits. The credit pool only applies to code that calls Claude programmatically.

Anthropic has confirmed that third-party agent tools, including OpenClaw, are being reinstated. After blocking access in April, they are now allowed back — but on the new metered model rather than as an effectively unlimited subscription perk.

Credits are billed at API list prices with no rollover. If you hit your monthly limit, usage stops until the next billing cycle unless you add credits separately.

Anthropic is sending email notifications to affected users on June 8 — one week before the change. Look out for that in your inbox.

What This Means for Business

For teams running light Claude usage — occasional API calls, developers using Claude Code a few times a day — neither change is disruptive. Swap the model ID, accept the new billing structure, move on.

For teams with heavier automated workloads, the math matters more. The $20 Pro credit covers roughly 4 million input tokens on Sonnet 4.6 at standard rates, which sounds like a lot until you are running large-context tasks through a pipeline daily. The Max plans give more room, and the API lets you purchase additional credits beyond the included pool.

The broader pattern here is worth noting. Over the past few months, Anthropic has been systematically moving Claude from an “all you can eat” subscription model toward usage-based metering for automated workloads. GitHub Copilot made the same shift on June 1. The era of flat-rate AI usage for production automations is closing, and that changes how you need to budget for AI infrastructure.

What to Do Before June 15

1. Export your usage data from the Claude Console and identify any calls to claude-sonnet-4-20250514 or claude-opus-4-20250514
2. Update model IDs in your code to claude-sonnet-4-6 or claude-opus-4-8
3. If you use Opus 4 and set temperature, top_p, or top_k explicitly, remove those parameters before upgrading
4. Test your migrated code thoroughly before June 15 — do not swap model IDs in production on the deadline day
5. Review how your team uses the Agent SDK, Claude Code automations, and third-party tools — audit what volume they generate so you can right-size your plan before the credits go live

If you need help thinking through how AI tools fit into your broader data and automation strategy, that is exactly what EDNA’s Omni Advisory service is designed for.