ClaudeHowSupport Us

claude-opus-4-6 — pricing, context window and limits

The last Opus with a 4096-token cache minimum and no xhigh effort level. It still accepts temperature/top_p/top_k and the deprecated budget_tokens escape hatch — every later Opus rejects all four with a 400.

Price per million tokens

Input$5
Output$25
Cache write, 5-minute TTL$6.25
Cache write, 1-hour TTL$10
Cache read$0.50 (0.1x input)
Batch input / output$2.50 / $12.50

Limits and behaviour

StatusCurrent
Context window1M tokens
Max output128K tokens
Min cacheable prefix4,096 tokens
Effort levelslow, medium, high, max
Effort defaulthigh
ThinkingOff unless requested (adaptive)
Forced tool_choiceAccepted
PlatformsClaude API, Amazon Bedrock, Google Cloud, Microsoft Foundry

Lifecycle

API model nameclaude-opus-4-6
Retires no sooner than5 Feb 2027
SuccessorClaude Opus 5.5

The last of the old request shape

Opus 4.6 is the final Opus that still accepts the older sampling controls (temperature, top_p, top_k) and the deprecated thinking-budget mechanism, and the last without the effort tier every later generation offers. Every Opus released after it rejects all four of those fields outright, which makes this model the boundary line for a whole category of migration work rather than one incremental step along it.

The highest cache minimum among active models

This model's minimum cacheable prefix sits at the top of the range across the active lineup. A prompt engineered to cache here will cache almost everywhere else too, but the reverse isn't guaranteed — and moving away from this model often means prompts that never cached here start caching on the destination with no code change.

What breaks first on a migration away

The older sampling fields usually surface first in testing, because removing them produces an immediate, hard error rather than a subtle behaviour change. See temperature, top_p and top_k are not supported on this model and budget_tokens is not supported on Opus 4.7 and later for both fields this model is the last to accept. The current migration target is Opus 5.5, reached through the Opus 4.7, 4.8 and 5 changes in order.

See migrating: Opus 4.6 to Opus 5 for the largest single diff, and what actually changed from Opus 5 to Opus 5.5 for the last step on the way to the current Opus.

Verified 2026-09-30 against claude-api skill — shared/models.md, model-migration.md + SKILL.md model table and Anthropic's pricing and model-deprecation pages (canonical: https://platform.claude.com/docs/en/about-claude/models/overview, https://platform.claude.com/docs/en/about-claude/pricing, https://platform.claude.com/docs/en/about-claude/model-deprecations).

Verified 2026-10-28 against Anthropic's model-deprecations status table and pricing page (canonical: https://platform.claude.com/docs/en/about-claude/model-deprecations, https://platform.claude.com/docs/en/about-claude/pricing).Could not confirm: These dates apply to the Claude API, Claude Platform on AWS, Claude Models on Google Cloud (Anthropic managed) and Microsoft Foundry. Amazon Bedrock and Claude Models on Google Cloud (Google managed) keep their own retirement schedules, which we do not track.