Claude Opus 5.5 vs Opus 5: Should You Upgrade?

Both keep a 1M-token window and 128K tokens of output

Opus 5 was released on July 24, 2026 and Opus 5.5 on September 22, 2026. Both have a 1M-token context window and a 128K-token output limit, or up to 300K on the Batch API with the output-300k-2026-03-24 beta header, so neither limit is a reason to move or to stay. The knowledge cutoff moves from May 2026 to June 2026.

Opus 5 stays available on the API as a legacy model, with retirement not sooner than July 24, 2027. Anthropic says a legacy model "will no longer receive updates and may be deprecated in the future", and its Opus 5 page tells users to "consider migrating to Claude Opus 5.5".

Every Opus 5.5 rate is lower, cache reads most of all

Prices are per million tokens on Anthropic's API. Input, output and cache writes fall by 20%. Cache reads fall by 60%, which matters most for agents that resend the same repository or document on every turn.

As an illustration, take one agent turn that reads 100K tokens from the cache, adds 10K new input tokens and writes 5K output tokens. On Opus 5 it costs $0.05 + $0.05 + $0.125 = $0.225. On Opus 5.5 it costs $0.02 + $0.04 + $0.10 = $0.16, about 29% less. Anthropic's 40% figure is measured at default settings, where Opus 5.5 also runs at a lower default effort. Customers quoted in its announcement report fewer output tokens, such as Factory's 20 to 25% fewer, so part of that saving likely comes from shorter runs at the lower default. If you set effort back to high, your saving may sit closer to the list-price difference.

Fast mode gives up to 2.5 times the output speed on both models at twice the standard price. It is a research preview on Anthropic's own API that needs access approval, and it is not offered on Bedrock, Google Cloud or Microsoft Foundry or with the Batch API. In Claude Code on a subscription it runs on usage credits only.

Anthropic's launch table: Opus 5.5 ahead on every row

These figures come from Anthropic's Opus 5.5 announcement, which ran Opus 5.5 at max effort except on Terminal-Bench 4.0 (xhigh). The table does not state Opus 5's setting. The largest gains are on agentic terminal and automation work.

Neither announcement states a SWE-bench Verified or GPQA Diamond score in its text, and the Opus 5 announcement shows its scores only as chart images. Treat such figures for this pair as unsourced unless they cite Anthropic.

What to change when you switch to claude-opus-5-5

• Thinking cannot be turned off. On Opus 5 you could disable thinking at high effort or below. On Opus 5.5, thinking: {"type": "disabled"} returns a 400 error at every effort level, so remove it.
• Forced tool use returns an error. Anthropic lists it among the breaking changes from Opus 5. Rewrite calls that force a particular tool before you switch them.
• The default effort is one level lower. Opus 5 defaults to high and Opus 5.5 to medium, so a request without an effort setting "runs one level lower". Set effort explicitly while you compare the two models.
• Leave room for thinking in max_tokens. Thinking tokens count toward max_tokens and are billed as output, even when the thinking text isn't returned.
• Unchanged: both models already reject budget_tokens, and any non-default temperature, top_p or top_k, with a 400 error, so a working Opus 5 request needs no change there.

Keep Opus 5 only while those calls are rewritten

Opus 5 is worth keeping for calls that switch thinking off for speed, and for pipelines that depend on forced tool use, until they are rewritten. If a fast answer without thinking is the point, Sonnet 5.5 can skip up-front thinking with between_tools at high effort or below, at $2 input and $10 output per million tokens; Claude Sonnet 5.5 vs Opus 5.5 covers that trade. You have until at least July 24, 2027.

In Claude Code there is nothing to migrate: it uses Opus 5.5 by default on Pro, Max, Team, Enterprise and the API, and Anthropic raised five-hour limits on Pro, Max, Team and seat-based Enterprise when Opus 5.5 launched. For the step above Opus 5.5, see Claude Opus 5.5 vs Fable 5.2 vs Sonnet 5.5.