DeadAI Dev Tools

Claude Sonnet 4 (claude-sonnet-4-20250514) and Claude Opus 4 (claude-opus-4-20250514)

by Anthropic

Announced

2026-04-14

Complete

2026-06-15

Refund window closes

—

Refund status

No refund applicable

On April 14, 2026, Anthropic notified developers using Claude Sonnet 4 and Claude Opus 4 that both would retire on the Claude API on June 15, 2026. That date has passed. Anthropic's deprecations page lists `claude-sonnet-4-20250514` and `claude-opus-4-20250514` as Retired, and says requests to retired models fail. The recommended replacements are `claude-sonnet-5-5` for Sonnet 4 and `claude-opus-4-8` for Opus 4. The date applies to Anthropic-operated platforms (the Claude API, Claude Platform on AWS and Microsoft Foundry). Anthropic's Sonnet 5.5 migration guide says Sonnet 4 is still available on Amazon Bedrock and Google Cloud, which set their own schedules. If an old script, notebook or pipeline step still names either model ID and calls the Claude API directly, it is now erroring.

Refund flow

1
Not applicable. Claude API usage is billed per token in arrears, with no balance tied to a model ID, so there is nothing to refund for the retirement itself.
2
If you see Claude API charges for claude-sonnet-4-20250514 or claude-opus-4-20250514 requests dated after June 15, 2026, treat it as a billing error and contact Anthropic support with the request IDs.
3
Re-budget using Anthropic's pricing page, read October 18, 2026. Opus 4 was listed at $15 per million input tokens and $75 per million output tokens. Opus 4.8 is $5 and $25, one third of the per-token price.
4
Sonnet 4 was listed at $3 and $15. Sonnet 5.5 is $2 and $10. The pricing page says Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text, while Sonnet 4.6 and earlier use the previous one. Both replacements are on the newer tokenizer, so on price and token count alone the same text costs roughly 13% less on Sonnet 5.5 and roughly 57% less on Opus 4.8. Thinking tokens are billed as output and can change that. Measure real requests before you assume a saving.

Migration path

Recommended successor

Claude Sonnet 5.5 (claude-sonnet-5-5) and Claude Opus 4.8 (claude-opus-4-8)

Why this is the closest fit

These are the replacements Anthropic names on its deprecations page: claude-sonnet-5-5 for claude-sonnet-4-20250514 and claude-opus-4-8 for claude-opus-4-20250514. Both are listed as Active, Sonnet 5.5 with retirement not sooner than September 28, 2027 and Opus 4.8 not sooner than May 28, 2027. Anthropic's Sonnet 5.5 migration guide covers moving from Sonnet 4 directly, with a checklist section for Claude Sonnet 4 or earlier.

What differs from the original

Per the Sonnet 5.5 migration guide, coming from Sonnet 4 you apply every section of the guide plus a Sonnet 4 section. Prefill of the last assistant turn returns a 400 error. thinking.type "enabled" with budget_tokens returns a 400 error; use adaptive thinking with output_config.effort. A request with no thinking field now runs adaptive thinking, so code that reads content[0].text can break; read content blocks by type. The Sonnet 4 section adds: use text_editor_20250728 (tool name str_replace_based_edit_tool, no undo_edit) and code_execution_20260521, handle the refusal and model_context_window_exceeded stop reasons, expect trailing newlines to be kept in tool call string parameters, and remove the token-efficient-tools-2025-02-19 and output-128k-2025-02-19 beta headers. For Opus 4 to Opus 4.8, no Opus 4.8 migration guide was read for this page, so no Opus specific breaking change is claimed. Test your own requests on it.

Alternatives by use case

Your code named claude-sonnet-4-20250514

Move to claude-sonnet-5-5 at $2 / $10 per million tokens and work through the Claude Sonnet 4 or earlier section of the migration checklist

You want the fewest behavior changes from Sonnet 4

claude-sonnet-4-6 keeps the older tokenizer and $3 / $15 pricing, does not think by default, and is listed as retiring not sooner than February 17, 2027. It still rejects prefill, so it buys time rather than a free swap

Your code named claude-opus-4-20250514

Move to claude-opus-4-8 at $5 / $25 per million tokens, the replacement Anthropic names, and re-run your evals

You call Sonnet 4 through Amazon Bedrock or Google Cloud

Anthropic says Sonnet 4 is still available there. Check your provider's own model table for its retirement date

What this tool meant

Claude Sonnet 4 and Opus 4 shared the 20250514 snapshot date and ran on the Claude API for about 13 months. Sonnet 4 in particular was a common default in scripts and tutorials, so the model ID lingers in code that nobody has opened since. For a video or content pipeline that uses Claude to draft prompts, shot lists or captions, a retired model fails loudly, which is the good case. The quieter risk comes after the fix: the replacements think by default, count tokens differently and reject prefill, so outputs and costs can shift even when every call succeeds. Run a fixed set of real inputs through the new model and compare text and token counts.