claude-opus-4-7 — pricing, context window and limits
Introduced the tokenizer every later Opus, Sonnet and Fable model still uses, the xhigh effort level, and high-resolution vision. Its cache minimum is 2048 tokens — higher than both the model before it and the model after it, which is the single most expensive thing to assume about this generation. Fast mode was removed: speed: "fast" on 4.7 errors.
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
| Status | Current |
|---|---|
| Context window | 1M tokens |
| Max output | 128K tokens |
| Min cacheable prefix | 2,048 tokens |
| Effort levels | low, medium, high, xhigh, max |
| Effort default | high |
| Thinking | Off unless requested (adaptive) |
| Forced tool_choice | Accepted |
| Platforms | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry |
Lifecycle
| API model name | claude-opus-4-7 |
|---|---|
| Retires no sooner than | 16 Apr 2027 |
| Successor | Claude Opus 5.5 |
The generation that introduced the current tokenizer
Opus 4.7 is where the tokenizer every later Opus, Sonnet and Fable model still uses was introduced, along with the effort tier just below the maximum and high-resolution vision. Anything built on this model or a successor inherits that tokenizer's counting, which matters the moment you compare token counts against an older Opus: the same text produces more tokens here than it did before.
The cache-minimum outlier
This model's minimum cacheable prefix is higher than both the generation before it and every generation after it — the most expensive thing to assume incorrectly about it if you think of cache minimums as trending steadily in one direction. See the minimum-cacheable-prefix table for the full, non-monotonic picture across the lineup.
Fast mode was removed here specifically
Unlike the Opus generations before and after it, this one does not support fast mode at all — requesting it is rejected. See speed: "fast" is not supported on Opus 4.7 for what that failure looks like and how to route around it.
Where to go next
The migration guide now names Opus 5.5 as the target for code on this model, reached by applying each generation's changes in order. The step to Opus 4.8 changes nothing in the request; the steps after it turn thinking on by default, then make it impossible to switch off, and remove forced tool choice.
Related
See migrating: Opus 4.7 to Opus 5 for the first diff, then migrating: Opus 5 to Opus 5.5 for the last.
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.