Independent publication. Not affiliated with or endorsed by Anthropic PBC.

Selected high-impact Claude API changes

Know what changed before it breaks production.

Independent, source-linked guidance on Claude model and API changes: what is affected, what to test, and what to do next.

Pilot boundary: selected operator-impacting changes verified Sep 11, 2026. This is not a complete or real-time feed.

Fable 5.1 rejects forced tool choice

A request with tool_choice set to any or a named tool returns a 400 invalid_request_error. Retrying the same body does not make it compatible.

Do this before cutoverReplace forced tool choice with auto plus an explicit instruction and a strict tool schema, or use structured outputs when the real requirement is schema-valid JSON.See exact impact and evidence

Three changes worth an operator decision

Open the full ledger →
Action required

Fable 5.1 rejects forced tool choice

Requests using tool_choice type any or tool now fail before inference; auto and none remain supported.

Surface
Claude Messages API · Message Batches API
When
Before sending production traffic to claude-fable-5-1
Check my impact
Monitor

Thinking-block compatibility is one-way

Fable 5.1 can read older-model thinking, but routing the same conversation back to an older model drops Fable 5.1 thinking blocks.

Surface
Multi-model routers · Client-side retries
When
Before enabling fallback or downgrade routing from Fable 5.1
Check my impact

Fable 5 → Fable 5.1, without optimistic assumptions

Messages API integrations that manage tools, conversation history, retries, fallback routing, or prompt caching.

Run the migration checklist
  1. Remove incompatible request controls.
  2. Keep conversation prefixes stable.
  3. Exercise downgrade routes.
  4. Re-baseline the workload before cutover.

6/6 frozen fixtures matched their expected classification

This checks request shape and conversation-prefix stability only. No Anthropic API request was made.

2 ready1 review3 blocked
Inspect the method, fixtures, and limits →

A claim earns its label.

Official facts, local observations, implications, and unknowns are kept separate. Material changes are reviewed against a 48-hour target and rechecked every 30 days.

Read the methodology

Twenty chapters of context

The existing cited history remains available as background. Current production decisions should start with the impact ledger.

Read the 2021–2026 history