Claude Opus 5.5

Anthropic’s own docs tell you to start here if you’re not sure which Claude model to use. It reads 1M tokens of input, writes up to 128K tokens, and decides for itself how much to think, which you steer with the effort setting (default medium).

When to choose it

• Multi-step work that runs for a long time: agents, refactors, or building a model or report that spans many files.
• Long documents such as filings, contracts or data rooms, where you want the whole set in one request. 1M tokens is about 555,000 words.
• You already use Claude apps or Claude Code and want the model most people there use.

Choose Claude Fable 5.1 ($10 / $50) instead only when Opus 5.5 at higher effort still falls short on your own tests. For high-volume or latency-sensitive jobs, Claude Sonnet 5 ($2 / $10) costs half as much.

Specs (from Anthropic, 2026-09-25)

API ID	claude-opus-5-5
Context window	1M tokens, same price per token across the whole window
Max output	128K tokens (up to 300K on the Batch API with a beta header)
Price	$4 per million input tokens, $20 per million output tokens
Discounts	Batch API 50% off; cache reads cost 5% of the input price
Knowledge cutoff	June 2026
Retirement	Not before September 22, 2027

Before you choose

Thinking is always on and you can’t turn it off; lower effort if you need faster, cheaper answers. Model IDs are pinned snapshots, so the model behind claude-opus-5-5 will not change under you.