[v5.0] reproduction: @ai-sdk/anthropic: messagestart for a different message while the

[ai-sdk-factory]

Bug

@ai-sdk/anthropic: message_start for a different message while the previous message is still open is silently merged into one generation, corrupting thinking blocks and failing all subsequent requests

Reproduction

• On release-v5.0, ai 5.0.228 and @ai-sdk/anthropic 2.0.93 accept msg_second while msg_first remains open without emitting InvalidResponseDataError.
• examples/ai-functions/src/reproduction/issue-18331-anthropic-spliced-message-start.ts observed both signed reasoning blocks in one completed streamText step, dropped unfinished toolu_first, and completed only toolu_second; expected behavior is an explicit stream error with no completed corrupted step.
• The same-ID duplicate emitted two response-metadata parts and replaced input-token usage from 10 with 99 instead of being ignored.
• Sequential message_stop followed by message_start remained valid, producing two finishes without errors.
• packages/anthropic/src/issue-18331.test.ts and fixtures under packages/anthropic/src/__fixtures__/issue-18331-*.chunks.txt preserve failing Node and Edge coverage for the malformed streams and passing coverage for sequential messages.
• The exact Google Vertex live splice and subsequent provider 400 were not exercised because Google Vertex live-service checks are skipped by policy; Anthropic documentation confirms tool-loop thinking blocks must remain complete and unmodified and that modified blocks receive HTTP 400.

Relevant Documentation

• https://platform.claude.com/docs/en/build-with-claude/streaming
• https://platform.claude.com/docs/en/build-with-claude/extended-thinking
• https://platform.claude.com/docs/en/api/errors

Related Issues

Reproduces #18331