Claude Opus 5.5: Specs, Pricing, Limits and Availability Facts

Claude Opus 5.5 is a proprietary large language model from Anthropic, released on 22 September 2026 as the first model in the Claude 5.5 family. This profile collects the facts Anthropic publishes about it: API ID, context window, output limit, prices, platforms, benchmarks, system card findings, data handling and retirement date. It is written for developers, architects, buyers and governance teams who need checked figures rather than summaries. Every fact comes from Anthropic's own documentation, announcement or system card, cited in the sources list. Where Anthropic does not publish a figure, the page says so.

Last checked: 2026-11-15.

This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider. It uses Anthropic's public documentation as of November 2026. Model details change, so confirm against Anthropic's current documentation.

Key takeaways

• Released 22 September 2026. Anthropic's release notes and announcement date the launch of claude-opus-5-5 to 22 September 2026, and the announcement calls it "the first model in our new Claude 5.5 family".
• 1M tokens in, 128K tokens out. The models overview lists a 1M-token context window, 128K max output tokens, text and image input and text output, with a reliable knowledge cutoff of June 2026.
• $4 input and $20 output per million tokens. Standard API prices are $4 and $20 per MTok, cache hits are $0.20, the Batch API halves the rates and fast mode costs $8 and $40.
• Thinking is always on. Opus 5.5 uses adaptive thinking that cannot be disabled, with a default effort of medium; thinking and forced tool_choice settings that worked on earlier models return a 400 error.
• Active, retirement not sooner than 22 September 2027. The deprecations page lists the model as Active, and Anthropic gives at least 60 days' notice before retiring a publicly released model.

What is Claude Opus 5.5?

Claude Opus 5.5 is the Opus-line model of Anthropic's Claude 5.5 family. The models overview describes it as "For long-running agentic coding and knowledge work" and rates its comparative latency as Moderate. It replaced Claude Opus 5 as the newest Opus model at a lower list price: $4 and $20 per million tokens against $5 and $25 for Opus 5. Anthropic's announcement states that Opus 5.5 costs 40% less to run than Opus 5 on typical workloads at default settings.

The model ID claude-opus-5-5 follows the undated naming used for Claude 4.6 and later models. Anthropic's versioning page states that such an ID "is not an alias. It is the snapshot": the weights behind it do not change, and an updated version would ship under a new ID. Serving infrastructure such as routers and safety classifiers can still change. For how Anthropic and other providers name versions, see LLM Model Naming and Versioning Explained. For the whole Claude line-up, see Anthropic Models.

Key facts at a glance

As of 2026-11-15. Source: Anthropic models overview, pricing page, model deprecations page and release notes.

Release and version history

As of 2026-11-15. Source: Anthropic release notes. Entries are listed by date, oldest first.

The launch notes list migration changes from earlier models. Sending thinking with type disabled or enabled returns a 400 error, so code should omit thinking and use the effort parameter. The tool_choice types any and tool return a 400 error; Anthropic directs users to auto with strict tool use. Computer use on the Claude API and Google Cloud requires the computer_toolset_20260801 toolset, while computer_20251124 still works on Amazon Bedrock.

How much does Claude Opus 5.5 cost?

As of 2026-11-15. Source: Anthropic pricing page. Prices in US dollars per million tokens (MTok).

Modifiers: setting inference_geo to "us" on the first-party API applies a 1.1x multiplier to all token categories. Regional and multi-region endpoints on Amazon Bedrock and Google Cloud carry a 10% premium over global endpoints for Claude 4.5 and later models. Claude Platform on AWS and Claude in Microsoft Foundry bill at the same per-model rates, converted to Claude Consumption Units at $0.01 per CCU. Caching and data residency multipliers stack with fast mode.

Tool costs: the tool use system prompt adds 286 tokens for auto or none; the pricing page lists no figure for any or tool on Opus 5.5. Web search costs $10 per 1,000 searches plus tokens. Code execution is free when a 20260209 or later web search or web fetch tool is in the request; otherwise each organisation gets 1,550 free hours a month, then $0.05 per container hour.

Context window and output limits

The models overview recommends the Models API as the more reliable source for programmatic checks, since it returns live capabilities and max_input_tokens values. Long prompts still cost input tokens at the standard rate, so a full 1M-token request costs $4 in input before caching.

Modalities and tools

• Input and output: text and image input, text output.
• Thinking: adaptive thinking, always on, controlled with the effort parameter (default medium). The system card and announcement report benchmark results at max effort unless stated otherwise.
• Tool use: tool_choice auto with strict tool use; any and tool return a 400 error.
• Computer use: computer_toolset_20260801 on the Claude API and Google Cloud; computer_20251124 on Amazon Bedrock.
• Server tools: web search, web fetch and code execution are priced on the pricing page; bash and text editor tools add 325 and 974 input tokens respectively for Claude 4.7 and later.
• Fast mode: generally available since 15 October 2026; the announcement describes it as up to 2.5x speed.

For security controls around agentic tool use, see the Claude Agent SDK Security Guide and the Claude Code Security Guide.

Where is Claude Opus 5.5 available?

As of 2026-11-15. Source: release notes and announcement.

Partner platforms set their own retirement schedules, so dates on Vertex AI and Bedrock can differ from the Claude API. For enterprise controls in the Claude apps, see the Claude for Enterprise Security Guide.

Access restrictions

Claude Opus 5.5 is generally available: the release notes list no access request for the base model, and fast mode has needed no access request since 15 October 2026. Usage is governed by the account's usage tier and rate limits. The system card describes safeguards that act on specific request types rather than on who can use the model:

• Cyber: a three-stage system (activation probe, lightweight classifier, LLM classifier); blocked requests fall back to Opus 4.8. Vulnerability discovery in source code is allowed, discovery in compiled binaries is blocked, and a Cyber Verification Program offers reduced restrictions to verified users.
• Chemical and biological: research biology classifiers, with blocks falling back to Opus 5.
• Frontier AI development: narrow classifiers, with blocks falling back to Opus 5.
• Conventional weapons and distillation: blocking classifiers with no fallback.

The system card states that blocks are transparent and do not covertly alter responses.

Benchmarks (provider-reported, dated)

Provider-reported scores from Anthropic's announcement of 22 September 2026, at adaptive thinking and max effort unless stated. Anthropic notes production safeguards were active during evaluation, which likely lowered some scores. Only Anthropic models are shown. Independent leaderboard figures are not included because none were fetched for this page.

Benchmark scores depend on effort setting, tools and harness, so they are not directly comparable with figures produced under other conditions.

Model card and system card

Anthropic published "System Card: Claude Opus 5.5", dated 22 September 2026. For what these documents contain in general, see Model Cards and System Cards Explained. The card states:

• Framework: it cites Anthropic's Responsible Scaling Policy, the Frontier Compliance Framework and the August 2026 Risk Report, and names California SB 53 and the EU AI Act General-Purpose AI Code of Practice as regimes the framework covers.
• Training data: a proprietary mix of public internet information, public and private datasets, user data shared through feedback or with permission, and synthetic data; the ClaudeBot crawler follows robots.txt. Knowledge cutoff June 2026.
• Chemical and biological risk: treated as CB-1 but not CB-2.
• Alignment: Anthropic assesses the risk of catastrophic harm from misalignment as "low"; reported regressions include following malicious instructions in pasted text and accepting unverifiable authorisation claims more often.
• Sections: RSP risk areas, cyber, safeguards and harmlessness, agentic safety including prompt injection, alignment, model welfare and capabilities.

See the Frontier AI Model Risk Frameworks Guide for how such reports are structured.

Data handling

• Zero data retention: the announcement states Opus 5.5 is available with ZDR. ZDR is enabled per organisation by Anthropic's sales team and covers eligible features on the Messages and Token Counting endpoints.
• Covered Models: Opus 5.5 is not on Anthropic's Covered Models list (Fable and Mythos models), which requires 30-day retention.
• Feature retention: Batch processing data is kept 29 days and code execution data up to 30 days; flagged content can be kept up to 2 years even under ZDR.
• Training use: the ZDR page states retained data is never used for model training without express permission.
• Data residency: inference_geo "us" keeps inference in the US at a 1.1x price multiplier.

For key handling around API access, see the Anthropic API Keys Security Guide.

Deprecation status

"Not sooner than" is the earliest possible date, not a fixed schedule.

How Claude Opus 5.5 differs from other models in its family

As of 2026-11-15. Source: models overview and pricing page. Models listed by release date, newest first.

Within the Opus line, Opus 5.5 lists lower input, output, cache and fast mode prices than Opus 5 ($0.20 against $0.50 for cache hits, $8 and $40 against $10 and $50 for fast mode). Within the 5.5 family, the Sonnet and Haiku models list lower prices and faster comparative latency. Release-by-release detail is on Claude Opus 5.5 vs Opus 5 vs Opus 4.8 and Claude Opus vs Sonnet vs Haiku 5.5.

Practitioner checklist

• Pin claude-opus-5-5 explicitly and record the ID in your model inventory with its "not sooner than" retirement date.
• Remove thinking disabled or enabled settings and forced tool_choice from code migrated from earlier models, and test for 400 errors before cut-over.
• Switch computer use integrations on the Claude API and Google Cloud to computer_toolset_20260801.
• Set the effort parameter deliberately; benchmark figures were reported at max effort while the default is medium.
• Model costs with caching, batch, fast mode and the 1.1x inference_geo multiplier included, not just the $4 and $20 list prices.
• Confirm ZDR enablement per organisation and note which features you use are outside ZDR.
• Read the system card's safeguards section so blocked cyber or biology requests falling back to another model are expected, not treated as faults.
• Subscribe to deprecation notices and check partner cloud schedules separately.

Standards and references

• Anthropic: Models overview (accessed 2026-11-15)
• Anthropic: Pricing (accessed 2026-11-15)
• Anthropic: Claude Platform release notes (accessed 2026-11-15)
• Anthropic: Model deprecations (accessed 2026-11-15)
• Anthropic: System Card, Claude Opus 5.5 (accessed 2026-11-15)
• Anthropic: Claude Opus 5.5 announcement (accessed 2026-11-15)
• Anthropic: Zero data retention (accessed 2026-11-15)
• Anthropic: Model IDs and versioning (accessed 2026-11-15)

Related NHI Mgmt Group resources: Anthropic Models, Claude Sonnet 5.5, Reasoning Models Explained, Claude for Enterprise Security Guide.