Claude Opus 5.5 is Anthropic's first model in the Claude 5.5 family, announced on September 22, 2026. Official API pricing is $4 per million input tokens and $20 per million output tokens—the same $4 / $20 figures that circulated in market chatter, confirmed on Anthropic's product page and the Claude Platform model overview.
Anthropic positions it for long-running agentic coding and knowledge work: stronger than Claude Opus 5 on its published suites, priced below Opus 5, and close enough to Claude Fable 5.1 on many workloads that the company says real-world gaps are narrower than the benchmark tables suggest.
Official Pricing Versus Claude Opus 5
From Anthropic's Claude Opus 5.5 announcement and the model overview (USD per million tokens):
| Item | Claude Opus 5.5 | Claude Opus 5 |
|---|---|---|
| Input | $4 | $5 |
| Output | $20 | $25 |
| Cache read | $0.20 | $0.50 |
| 5-minute cache write | $5 | $6.25 |
| Batch API | 50% off input and output | (same discount pattern on the pricing page) |
Anthropic states that at default settings Opus 5.5 costs about 40% less than Opus 5 on typical workloads, combining the lower per-token rates with fewer tokens per task. Cache reads are 5% of base input ($0.20), versus 10% on most Claude models and 2.5% on Fable 5.1 / Mythos 5.1.
Fast mode (Claude Code and Claude Platform) is priced separately at $8 input / $40 output per million tokens, with up to about 2.5× speed claimed on the announcement page. One-hour cache writes are listed at $8 / MTok on the overview.
Specs and Model IDs
Per the Opus 5.5 overview:
| Spec | Value |
|---|---|
| Claude API model ID | claude-opus-5-5 |
| Context window | 1M tokens |
| Max output | 128K tokens (up to 300K on Message Batches with beta header output-300k-2026-03-24) |
| Thinking | Adaptive, always on (default effort medium) |
| Knowledge / training cutoff | June 2026 |
| Status | Active (latest); retirement not sooner than September 22, 2027 |
Platform IDs: Claude API / Google Cloud / Microsoft Foundry / Claude Platform on AWS use claude-opus-5-5; Amazon Bedrock uses anthropic.claude-opus-5-5. Anthropic lists availability across AWS, Google Cloud, and Microsoft Azure.
Versus Opus 5 and Claude Fable 5.1
On Anthropic's published table, Opus 5.5 leads Opus 5 across agentic coding, knowledge-work, and computer-use benchmarks in the announcement, and often sits near or ahead of Claude Fable 5.1 while charging $4 / $20 against Fable 5.1's $10 / $50. Anthropic's own note: at this capability level, benchmark margins are a weaker guide than day-to-day use, and the Opus 5.5 vs Fable 5.1 gap in their internal use is narrower than scores imply.
Claude Fable 5.5 is not a released Anthropic product. https://www.anthropic.com/claude-fable-5-5 returned 404 when checked for this article. The current Fable line in Anthropic's docs is Claude Fable 5.1 (claude-fable-5-1). Do not treat search chatter about “Fable 5.5” as the same model as Opus 5.5. For Fable pricing and migration detail, see Claude Fable 5.1. Older Opus pricing and rankings remain on Claude Opus 4.6; Gemini Argon comparisons sit on Argon vs Opus.
Migration From Claude Opus 5
The Opus 5.5 migration guide lists four request-level breaking changes for code still on Opus 5:
- Thinking cannot be disabled.
thinking: {"type": "disabled"}andthinking: {"type": "enabled", "budget_tokens": N}return HTTP 400. Drop thethinkingfield; steer depth withoutput_config.effort(for example"low"). - Forced tool use is rejected.
tool_choicetypesanyandtoolreturn 400. Useauto, preferably with strict tool schemas, and state the tool in the prompt. - Thinking blocks are bound to the model and conversation. On the Claude API, Fable 5.1 and Mythos 5.1 can read Opus 5.5 thinking blocks; other models generally cannot. Keep conversations append-only. For API accounts created on or after August 31, 2026, 00:00 UTC, preserved-thinking rules match Fable 5.1: editing earlier turns and replaying thinking blocks can 400.
computer_20251124is rejected on the Claude API and Google-managed Claude Models on Google Cloud. Switch tocomputer_toolset_20260801. On Amazon Bedrock,computer_20251124still works with Opus 5.5.
A non-failing shape change: text between tool calls arrives in thinking blocks whose text is empty when thinking.display stays at the default "omitted". Set a display mode that returns updates if your UI streams progress between tools.
Safety, Safeguards, and Access
Anthropic reports Opus 5.5 as its strongest model to date on its automated behavioral audit, with lower rates of hard-to-reverse actions and better prompt-injection resistance than Opus 5 in the tests it published. Because capability in biology and cybersecurity is described as comparable to Claude Mythos 5.1, deployment uses safeguards similar to Claude Fable 5.1: many cyber tasks fall back to Opus 4.8; biology work that hits safeguards needs the Life Sciences Verification Program; a Cyber Verification Program expansion for Opus 5.5 is described as forthcoming. Distillation protections include preserved thinking for eligible API accounts. Zero data retention is available as with prior Opus models. Thinking mode cannot be switched off.
Claude Sonnet 5.5 and Claude Haiku 5.5 are listed as following in the weeks after the Opus 5.5 launch. Separately, Anthropic's research-preview agent app is covered in Claude Duet—a product launch, not a model ID.
Rankings on AI Rank
On the Arena overall board published October 15, 2026 (fetched into AI Rank October 17, 2026), Claude Opus 5.5 (High) sits at No. 3 with rating 1508.2 over 8,300 votes—ahead of Claude Fable 5 (High) and Claude Opus 5 (High/Max) on that snapshot.
→ Track it on AI Rank: Claude Opus 5.5 (High) — Arena rating, rank and history
On AI Rank's OpenRouter text usage ranking for October 10–16, 2026, Claude Opus 5.5 (non-batch) ranked No. 5 with about 5.08T tokens under model id anthropic/claude-opus-5.5-20260921. OpenRouter's public catalog also lists anthropic/claude-opus-5.5 at $4 / $20 per million tokens (batch SKU half price). Upstream OpenRouter display names that append “(batch)” to the non-batch row are naming quirks; this article follows Anthropic's non-batch product name.
→ Live usage: see the OpenRouter model usage ranking on AI Rank (tokens, Oct 10–16, 2026 window).
Compare before you choose: pricing changes and rankings move daily — check the current model rankings on AI Rank before you commit.
Or 📱 get the AI Rank app to follow them on your phone.
Prepared with automated editorial assistance and checked against official Anthropic pages and dated AI Rank data on October 22, 2026 (America/Los_Angeles). We did not run our own benchmarks for this article.