> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ai-stats.phaseo.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Anthropic: Claude Sonnet 4.5 to 4.6

> The concrete API and behavior changes to review when upgrading from Claude Sonnet 4.5 to Claude Sonnet 4.6.

# Anthropic: Claude Sonnet 4.5 to 4.6

Use this guide when you are moving from `anthropic/claude-sonnet-4.5` to `anthropic/claude-sonnet-4.6`.

Sonnet 4.6 shares the same migration mechanics as the broader Claude 4.6 line, but it also changes the performance profile enough that prompt and routing assumptions should be revalidated.

## What changed

* assistant prefills are no longer supported on Claude 4.6 models and return `400`
* structured output moved from `output_format` to `output_config.format`
* adaptive thinking is now configured through `output_config.effort`
* outdated beta headers should be removed
* Sonnet 4.6 is a full upgrade across coding, computer use, long-context reasoning, agent planning, and instruction following
* Sonnet 4.6 adds a 1M-token context window in beta

## What to change in your integration

### 1. Stop using assistant prefills

If your existing Sonnet 4.5 prompts start the assistant with a JSON prefix, XML tag, or formatting stub, remove that pattern before rollout.

### 2. Move structured output config

Replace `output_format` usage with `output_config.format` and rerun any schema or parser tests.

### 3. Re-tune effort and latency expectations

If you use Sonnet for interactive routes, benchmark different `effort` levels before making 4.6 the default. The model is stronger, but the best setting depends on your latency budget.

## What to test

### Output behavior

* instruction-following on formatting-sensitive prompts
* agent planning quality on multi-step tasks
* coding, refactor, and review prompts
* long-context retrieval and summarization

### Contract stability

* JSON and schema validation pass rate
* tool argument completeness
* finish reasons and empty-output handling
* routes that formerly relied on prefills for stable formatting

### Production metrics

* latency by effort level
* task success versus Sonnet 4.5
* timeout and retry rates
* cost per successful task

## Safe rollout

1. Remove prefills and update structured output config first.
2. Evaluate Sonnet 4.6 on your current production prompt set.
3. Canary traffic with Sonnet 4.5 or another stable route as fallback.
4. Increase traffic only after quality and contract stability hold.

## Sources

* [Anthropic migration guide](https://platform.claude.com/docs/en/about-claude/models/migration-guide)
* [Introducing Claude Sonnet 4.6](https://www.anthropic.com/news/claude-sonnet-4-6)