Docs · Models

Claude Opus 5.5 special offer

2026-10-29

A temporary, low-priced Claude Opus 5.5 line that is often unavailable: prices, how to create a key for it, the extra tokens it adds and its terms.

The special offer is a separate, temporary line for claude-opus-5-5 at a much lower fixed price than the shared pool and official line. It uses the same endpoint, the same Anthropic- and OpenAI-compatible API and the same model id. Only an API key created on this line uses it, and the line comes with conditions the other routes do not have. Read them before you move work onto it.

This line is often unavailable, sometimes for long stretches, and its availability is not guaranteed. The low price comes from an upstream that frequently has no capacity. Do not build anything on it that has to work at a given moment.

Terms at a glance

Often unavailable. Expect long periods when every request fails. Requests can also slow down or be refused, and keys on this line may stop working. There is no fallback to another route, and availability is not guaranteed.
Temporary. The line can be withdrawn at short notice.
Extra prompt tokens. The line adds its own system-prompt text to every request to raise the cache hit rate. Those tokens appear in your usage and are billed, mostly as cache reads.
What is guaranteed. Your requests are answered by the official Claude Opus 5.5 model, and you are billed at the prices below for the usage recorded in your usage log.
What is not guaranteed. Response quality, speed, availability, feature support and the overall experience may differ from Anthropic's API and from our other Claude lines.

Price

USD per 1M tokens:

Lane	Special offer	Shared pool	Official line	Anthropic list price
Input	$0.45	$2.00	$3.60	$4.00
Output	$0.90	$10.00	$18.00	$20.00
Cache read	$0.10	$0.10	$0.18	$0.20
Cache write (5 min)	$0.20	$2.50	$4.50	$5.00
Cache write (1 h)	$0.30	$4.00	$7.20	$8.00

Input, output and cache writes cost a fraction of the other lines, and cache reads cost the same as on the shared pool. The line also adds prompt tokens of its own (see below), so if your requests are almost entirely cache reads with very little output, compare the Cost column of your usage log with the shared pool before moving that traffic.

There is no long-context tier. Prices are fixed in USD and charged from your prepaid balance like every other route (how requests are billed).

Create a key for this line

An existing key does not switch to this line by itself, and an Auto route key never chooses it.

Sign in, open API keys and click Create API key. The Create a key on this line button on the special-offer card of the home page opens the same form with the line already selected.
Under Group, find the Claude section and choose Special offer, marked Opus 5.5 only · limited time · often unavailable. It is listed only while the line has a working channel behind it. Expand its model-count button to see the price table.
The form lists the terms of the line below the group list. Read them; creating the key means you accept them.
Set a Quota (USD) if you want a spending cap, then click Save changes and copy the key (it starts with sk-).

To move an existing key instead, open it with Edit and change Group to Special offer. Keep a second key on the shared pool or official line, so you can switch back at once if this line stops serving.

Send requests to the usual endpoint with model claude-opus-5-5:

curl https://ai.topxea.com/v1/messages \
  -H "content-type: application/json" \
  -H "x-api-key: sk-xxxx" \
  -H "anthropic-version: 2023-06-01" \
  -d '{"model":"claude-opus-5-5","max_tokens":256,"messages":[{"role":"user","content":"Say hello in one sentence."}]}'

For Claude Code, use the key of this line and pin every model variable to Opus 5.5. This line sells no other model, so a switch to Sonnet or Fable would be refused:

export ANTHROPIC_BASE_URL=https://ai.topxea.com
export ANTHROPIC_AUTH_TOKEN=sk-...                    # a key on the Special offer line
export ANTHROPIC_MODEL=claude-opus-5-5
export ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-5-5
export ANTHROPIC_DEFAULT_SONNET_MODEL=claude-opus-5-5
export ANTHROPIC_DEFAULT_HAIKU_MODEL=claude-opus-5-5

Any other model on a key of this line gets HTTP 403 "Model is not available on this API key route", before anything is forwarded or charged.

The extra system prompt

The line puts its own system-prompt text in front of your request. It is the same for every request, so after the first request it is mostly served from the prompt cache. That raises the cache hit rate and keeps the line cheap, but the tokens are real usage: they are counted and billed like your own, as cache reads, and as cache writes when the cache is cold.

How many tokens the line adds is not fixed. In our tests, a one-sentence request on 2026-10-29 was billed 34 input and 8 output tokens, with no cache tokens. On 2026-10-28, a prompt that the line's own token counter put at 1,153 tokens was billed 4 input, 310 cache-write and 2,722 cache-read tokens, nearly 1,900 more than it counted, most of them as cache reads. The added text can change and cannot be switched off, so judge the line by the token counts in your own usage log.

The model reads these instructions too, so an answer can differ from the same request sent on the shared pool or the official line.

Stability and failures

This line is often unavailable. On its first day, 2026-10-29, the upstream stopped serving six minutes after the line opened and was still answering every request with an error more than half an hour later. Outages like this are expected and can last much longer; there is no availability guarantee. Errors, slow responses and refusals also happen while it does serve. There is no fallback: a failed request on this line is not retried on another Claude route.

During an outage a request gets HTTP 503 with the error type overloaded_error and "The service is temporarily unavailable. Please retry or contact support." Retry later, or send the request with a key on the shared pool or the official line.

A request that fails before producing output is not charged; the reservation is released and the usage log shows an error row.
A stream that breaks off after producing output is charged for the output produced so far, as on every route (common error responses).
Keys on this line may stop working, for a while or when the offer ends. Requests then fail without a charge. Move the key to the shared pool or the official line under Edit, or use your second key.

What we guarantee and what we do not

We guarantee that requests on this line are answered by the official Claude Opus 5.5 model, and that you pay the listed prices for the usage recorded in your usage log.

We do not guarantee an experience like the official API or our other lines. In particular, response quality and consistency, latency and throughput, availability, rate limits and support for every request option of the official API may differ. Use the shared pool or the official line for production work that needs predictable behaviour.

The Acceptable Use Policy covers this line like every other route. Questions go to support@topxea.com or the help button at the bottom right.

Models: claude-opus-5-5 · Models & pricing

Previous: Kimi K3.1 Pro · Next: Top up with a card

Other languages: 简体中文