Claude Opus 5.5 API: $4/$20 Pricing and Setup

Claude Opus 5.5 is Anthropic's Opus-tier model released 22 September 2026. The API id is claude-opus-5-5. Claude Opus 5.5 API pricing is $4 per 1M input tokens and $20 per 1M output tokens, with cache reads at $0.20 per 1M.

Opus 5.5 is not on RouterPlex yet. Create a RouterPlex account for the current 60+ models. Claude Opus 5 is already live at $5 / $25, and Claude Fable 5 is live at $10 / $50. Opus 5.5 joins that key once the marketplace lists it: vendor list price, 0% markup, $0 top-up fee, prepaid balance, hard spend limit.

Claude Opus 5.5 API pricing #

USD per 1M tokens. The announcement's cache-write cell of $5 is the 5-minute rate. The overview also lists a 1-hour cache write at $8.

A 40,000-in / 2,000-out turn with no cache is $0.20 on the Opus 5.5 sheet ($0.16 input + $0.04 output). The same shape on live RouterPlex claude-opus-5 at $5 / $25 is $0.25.

Anthropic's batch discount is 50% off input and output on the Anthropic platform. The minimum cacheable prompt is 512 tokens. Fast mode, in Claude Code and the Claude Platform, is up to 2.5x speed at $8 / $40 per 1M. Anthropic calls that a research preview on the Claude API. It is not a RouterPlex rate, and it is not something a third-party gateway bills unless that gateway lists the SKU.

Anthropic's own cost claim: about 40% less than Opus 5 on typical workloads at default settings, from token rates that are 20% lower plus fewer tokens, and cache reads 60% lower ($0.20 vs $0.50). Output is more than 30% faster than Opus 5. Those are vendor tests.

Specs and breaking changes #

Four breaks versus Opus 5, from the overview. Thinking cannot be disabled. Forced tool use returns an error. Thinking blocks are tied to the model and the conversation. On the Claude API and Google Cloud, computer_20251124 is not accepted. The first three also apply on Fable 5.1. Text between tool calls comes back in thinking blocks whose text is empty at the default display setting.

Sonnet 5.5, at $2 / $10, shipped on 28 September 2026 as the everyday companion. Haiku 5.5 was described as coming in the following weeks on the 22 September Opus page. Anthropic has not published a Haiku 5.5 price yet.

Vendor benchmarks #

Unless noted, Anthropic reports adaptive thinking at max effort. Terminal-Bench 4.0 is Opus 5.5 at xhigh and GPT-6 Astra at high. Safeguards were on: cyber tasks fell back to Opus 4.8, and biology or frontier-LLM tasks fell back to Opus 5, which Anthropic says likely lowers these scores.

Terminal-Bench 4.0 standard error is ±2.6 points for Opus 5.5. The public leaderboard lists Opus 5 at 51.8%. Anthropic's setup reproduced 52.3%. AutomationBench was run by Zapier without fallback models, so safeguard hits counted as failures. Terminal-Bench-Science standard error is ±3.5 to 5 points.

More vendor claims from the same announcement, each one Anthropic's or the named customer's: a 680,000-line migration in under a day. A load-time test at 39 of 40 pages. A 200,000-line audit in under 3 hours, where Opus 5 took over 20 hours and 2.5x the tokens. An internal HAProxy C-to-Rust port in 9.5 hours versus 12 hours for Fable 5.1, at 51% less cost. An earnings-research test where 16 of 18 Opus 5.5 reports cleared the bar. A fictional merger model in 63 minutes versus 93 for Opus 5, at 50% less cost. At default effort on FrontierCode, Anthropic says it beats GPT-6 Astra at roughly 20% of the cost per task. The announcement gives that as a ratio, not a dollar figure.

Mario Rodriguez, GitHub's Chief Product Officer, said Opus 5.5 used among the fewest tokens and steps in Copilot CLI and VS Code, and in VS Code solved more terminal tasks than Opus 5 in less than half the steps. Carl Bennett, CIO at Deloitte Consulting LLP, cited 72% of known bugs caught at the lowest effort setting versus 56% for Opus 5 at high effort. Gray Swan, named by Anthropic, tied Opus 5.5 with Fable 5.1 for the lowest prompt-injection success rate in that firm's test. Frontier Design and METR are named as evaluators on the system card.

Call Opus 5 today #

claude-opus-5 is the live RouterPlex model at $5 / $25 per 1M. Opus 5 fast mode at $10 / $50, documented on the Opus 5 guide, is first-party and is not this gateway's SKU.

curl https://api.routerplex.com/v1/chat/completions \
  -H "Authorization: Bearer $ROUTERPLEX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-opus-5",
    "messages": [
      {"role": "user", "content": "List the three risks in moving this service from Opus 5 to a newer id."}
    ]
  }'

Same key, OpenAI-compatible endpoint, full catalog.

On RouterPlex, once Opus 5.5 is listed #

The catalog id will be whatever the row is named. Expect vendor list price, 0% markup, prepaid, and a hard per-key budget that stops the request at $0. Register now for the 60+ models already on the marketplace, including Opus 5 and Fable 5, and use claude-opus-5-5 here when the row is live.