Anthropic Takes Opus 5.5 Fast Mode Out of Research Preview, Cutting the Waitlist for a 2.5x-Faster Tier

Anthropic has taken fast mode for Claude Opus 5.5 out of research preview. As of October 15, 2026, requests to claude-opus-5-5 no longer need the fast-mode-2026-02-01 beta header or a waitlist approval to run at the faster inference setting, according to the Claude Platform release notes. It is a small-looking line in a long changelog, but it removes the two concrete friction points — the waitlist and the beta header — that applied to Opus 5.5 fast mode since it launched in research preview in September.

What actually changed

Fast mode runs the same Opus 5.5 weights at a faster inference configuration: no change to intelligence or capabilities, just more output tokens per second. Anthropic’s fast mode documentation puts the gain at up to 2.5x higher output tokens per second (OTPS) compared with standard speed, with the benefit concentrated in streaming throughput. You opt in by setting speed: "fast" on a Messages API request.

The two concrete changes that landed this week:

• Requests to Claude Opus 5.5 no longer require the fast-mode-2026-02-01 beta header. Requests that still send the header keep working, so nothing breaks if you leave it in.
• Claude Opus 5.5 now has a default fast mode rate limit baked into the Start, Build, and Scale usage tiers. You no longer have to request access before you can test it; organizations still on the Evaluation tier need to move to Start, Build, or Scale before fast mode becomes available to them.

Fast mode for Opus 5.5 is also now available on Claude in Microsoft Foundry for deployments hosted on Azure, with Google Cloud support described as coming soon in a same-day post from the ClaudeDevs account on X.

The price of speed

Fast mode is not free speed. It is priced as a multiplier on standard rates across the full context window:

Model	Standard price	Fast mode price	Access
Claude Opus 5.5	$4 / $20 per MTok	$8 / $40 per MTok	Generally available, no waitlist
Claude Opus 5	$5 / $25 per MTok	$10 / $50 per MTok	Research preview, requires account manager or waitlist approval, plus the beta header
Claude Opus 4.8	standard pricing unchanged	$10 / $50 per MTok	Research preview, requires account manager or waitlist approval, plus the beta header

That is a 2x price premium over Opus 5.5’s standard rate for up to 2.5x throughput — the same advertised ceiling Anthropic lists for Opus 5 and Opus 4.8 fast mode, whose $10/$50 pricing sits on top of their higher standard rates. What changed isn’t the advertised speed-to-price ratio; it’s access. Opus 5.5 fast mode is now open to any account in a paid usage tier, while Opus 5 and Opus 4.8 fast mode still require approval from an account manager or the waitlist, plus the beta header. For a long-running agent loop that is bottlenecked on output generation rather than reasoning quality, paying 2x for up to 2.5x throughput can still shrink wall-clock time on a task enough to be worth it, but that is a claim to verify against your own workload, not a guaranteed return.

Fast mode pricing stacks with other multipliers: prompt caching discounts and data residency surcharges both still apply on top of the fast mode rate. And switching speed mid-conversation invalidates the prompt cache entirely, since requests at different speeds don’t share cached prefixes. If your workload leans on cache hits for cost control, toggling fast mode on and off turn by turn will cost you more than the headline multiplier suggests.

Where it does not help

Fast mode is still unavailable with the Message Batches API, incompatible with a Priority Tier commitment, and not currently offered on Claude Platform on AWS. Its benefits are also concentrated in output tokens per second rather than a promised improvement to time to first token, per Anthropic’s documentation, so a workload dominated by a long prompt and a short answer may see less benefit than one dominated by long streamed output.

What it means for Claude Code

Fast mode inside Claude Code itself is unchanged by this release. Per the ClaudeDevs post, /fast in Claude Code keeps drawing from extra usage credits rather than the plan’s standard usage pool, separate from the Claude Platform API change described above. For teams running Claude Code against subscription plans, that means the economics of flipping on /fast for a long agent session are still a usage-credit decision, not a new unlocked default. The platform-level change matters most to teams calling the Messages API directly, including through Claude Managed Agents, where speed: "fast" is now something you can flip on for Opus 5.5 and measure, instead of filing a request and waiting.

The practical takeaway

For anyone building or operating coding agents on top of Opus 5.5, the move removes a concrete barrier to testing fast mode: you no longer need to request access or wait for approval before measuring it against your own workload. Fast mode still runs on a dedicated rate limit separate from standard Opus capacity, so a request can still return a 429 if that limit is exceeded or a 529 if fast mode capacity is unavailable — plan retry or fallback logic for both cases. Within that constraint, write down what a 2x price increase buys you in wall-clock time for your specific agent loop before leaving speed: "fast" on by default, especially since it also forfeits prompt cache reuse across speed switches. Our related guide on writing a speed budget before anyone types /fast and on pricing effort per completed task both cover how to make that call without just eyeballing the invoice at the end of the month.

Fast mode for Opus 5 and Opus 4.8 remains gated in research preview at $10/$50 per MTok, with account manager or waitlist approval and the beta header still required, so the GA treatment so far is specific to Opus 5.5.

Load X post

Loading this post connects to X.

The original source is available above.