Claude Opus 4.8 Fast Mode Ends November 30: Moving to Opus 5.5 Cuts the Rate 20%
Anthropic removes fast mode from Claude Opus 4.8 on November 30, 2026. After that date, a request to claude-opus-4-8 with speed set to "fast" returns an error instead of running. The documented replacement is fast mode on Claude Opus 5.5, which lists at $8 input and $40 output per million tokens, 20% below the $10 and $50 that Opus 4.8 fast mode costs. Opus 4.8 itself stays available at standard speed. Only the fast option goes away.
Key facts
- October 30, 2026: Anthropic's API release notes deprecated fast mode for Claude Opus 4.8, with removal on November 30, 2026.
- After November 30, requests to claude-opus-4-8 with speed "fast" return an error. Anthropic's instruction is to migrate to Claude Opus 5.5 to keep using fast mode.
- Fast mode prices per million tokens: Opus 5.5 at $8 input and $40 output; Opus 5 and Opus 4.8 at $10 input and $50 output.
- Standard prices per million tokens: Opus 5.5 at $4 input and $20 output; Opus 4.8 at $5 input and $25 output.
- October 15, 2026: fast mode on Opus 5.5 left research preview. It needs no beta header and has a default rate limit in the Start, Build and Scale usage tiers.
What breaks on November 30
Fast mode runs the same model with a faster inference configuration. Anthropic's documentation puts the gain at up to 2.5x higher output tokens per second, with the benefit on output speed rather than time to first token. On Opus 4.8 it was a research preview that needed the fast-mode-2026-02-01 beta header.
The removal does not fall back quietly. Requests that keep asking for fast mode on Opus 4.8 fail, so any code path that sets speed "fast" on that model ID will start returning errors on December 1 unless you change it. If your application treats any API error as fatal, users see a failed request where they used to get a fast answer.
You have two exits. Drop the speed parameter and stay on Opus 4.8 at standard speed and standard price, or move the model ID to claude-opus-5-5 and keep fast mode.
What the move to Opus 5.5 costs
On list prices, the move is a cut. Take a workload of 50 million input tokens and 5 million output tokens a month in fast mode, with no caching. On Opus 4.8 that is $500 input plus $250 output, $750 in total. On Opus 5.5 it is $400 plus $200, $600 in total, a saving of $150 a month.
Caching widens the gap. Anthropic's pricing page says prompt caching multipliers apply on top of fast mode prices, and Opus 5.5 charges cache hits at 0.05x the base input price against 0.1x on most other models. Applied to the fast mode input rates, that works out to $0.40 per million cached tokens on Opus 5.5 against $1.00 on Opus 4.8. The minimum cacheable prompt also drops from 1,024 tokens on Opus 4.8 to 512 on Opus 5.5.
The saving per token is not the whole bill. On Opus 5.5 adaptive thinking is always on, and Anthropic's migration guide says thinking tokens are billed as output tokens, so workloads that ran without thinking on Opus 4.8 can produce more output tokens per request. The default effort level is medium on Opus 5.5 against high on Opus 4.8. Measure your real token counts after the switch before you book the 20%.
The request changes you have to make
Changing the model ID is not enough on its own. The Opus 5.5 migration guide lists settings that now return a 400 error:
- thinking set to "disabled", and manual thinking budgets with budget_tokens. Use the effort parameter instead.
- temperature, top_p and top_k. Omit them.
- A prefilled assistant turn at the end of messages. Use structured outputs or system prompt instructions.
Two more points matter for capacity planning. Priority Tier is not supported on Opus 5.5, while Opus 4.8 keeps it. Fast mode is not available with the Batch API or a Priority Tier commitment on any model. Falling back from fast to standard speed is a prompt cache miss, because requests at different speeds do not share cached prefixes.
What to do before November 30
- Search your code, config and prompt tooling for claude-opus-4-8 used together with speed "fast" or the fast-mode-2026-02-01 header.
- Decide per workload: standard speed on Opus 4.8, or fast mode on Opus 5.5.
- For the Opus 5.5 path, remove the rejected parameters, set effort explicitly, raise max_tokens to cover thinking plus the answer, and read content blocks by type.
- Run your own evals and cost baseline at the effort level you pick, then switch traffic before December 1.
- If you depend on Priority Tier, plan that capacity separately.
The same month-end date applies to OpenAI's GPT-5.6 Sol promo pricing, so teams running both vendors should review both bills together. For a longer migration window on another vendor, see our note on the Mistral Large 3 deprecation. More pricing analysis is in the article archive.
Questions people ask
Is Claude Opus 4.8 being retired?
No. Only fast mode on Opus 4.8 is removed on November 30, 2026. Requests without speed "fast" keep working at the standard $5 input and $25 output per million tokens.
What happens to a fast mode request on Opus 4.8 after November 30?
It returns an error. Anthropic's release notes and fast mode documentation both state this, and point to Claude Opus 5.5 as the model to use for fast mode.
Is Opus 5.5 fast mode cheaper than Opus 4.8 fast mode?
Per token, yes: $8 and $40 per million input and output tokens, against $10 and $50. Total cost depends on how many thinking tokens your requests produce on Opus 5.5, since thinking is always on and billed as output.
Do I need a beta header for fast mode on Opus 5.5?
No. Fast mode on Opus 5.5 left research preview on October 15, 2026. Opus 5 and Opus 4.8 still need the fast-mode-2026-02-01 header while their fast mode lasts.
Sources