Claude Opus 5.5, Opus 5 and Opus 4.8 are the three Opus models Anthropic released in 2026, and all three are still callable on the Claude API, so teams have to decide which one to pin and when to move. This page compares them field by field: API IDs, release dates, lifecycle status, context window, maximum output, knowledge cutoff, prices including cache, batch and fast mode, documented breaking changes, safeguards, provider-reported benchmarks and earliest retirement dates. It is written for developers migrating between Opus versions, buyers estimating cost and architects planning model lifecycles. Models are listed by release date, newest first, an objective order that implies no ranking. Every fact comes from Anthropic's own documentation and announcements, cited in the sources list.
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
- All three are Active. Anthropic's deprecations page lists claude-opus-5-5, claude-opus-5 and claude-opus-4-8 as Active, with earliest retirement dates of 22 September 2027, 24 July 2027 and 28 May 2027.
- Opus 5.5 is the only one priced below $5/$25. Opus 5.5 lists $4 input and $20 output per million tokens with a $0.20 cache read; Opus 5 and Opus 4.8 both list $5, $25 and $0.50.
- Same window, same tokeniser. All three list a 1M-token context window and 128K maximum output at standard pricing, and all three fall under Anthropic's "Claude 4.7 and later" tokeniser note, so per-token prices compare directly.
- Thinking control narrows with each release. Opus 4.8 rejects non-default sampling parameters; Opus 5 allows thinking to be disabled only at effort high or below; on Opus 5.5 thinking cannot be disabled and forced tool use returns a 400 error.
- Opus 4.8 fast mode ends on 30 November 2026. Anthropic deprecated fast mode for Opus 4.8 on 30 October 2026 and names Opus 5.5 as the migration path; fast mode for Opus 5 remains in research preview.
What is Claude Opus?
Claude Opus is one of five Claude model lines in Anthropic's documentation, alongside Fable, Mythos, Sonnet and Haiku. Anthropic's models overview describes Opus 5.5 as "for long-running agentic coding and knowledge work", rates its latency as Moderate and prices it above Sonnet 5.5 and Haiku 5.5. The individual model pages mark Opus 5 and Opus 4.8 as legacy models that remain available, with Anthropic recommending migration to Opus 5.5. Every Opus model is proprietary: Anthropic publishes no weights and states no parameter count. All three use undated API identifiers.
The three releases came roughly two months apart. Anthropic's release notes call Opus 4.8 "our most capable model" at its launch in May 2026, describe Opus 5 as "a step-change improvement over Claude Opus 4.8" in July, and introduce Opus 5.5 in September as the first model of the Claude 5.5 family. For the full Claude line-up, see the Anthropic Models hub, and for the single-model deep dive, the Claude Opus 5.5 profile.
Every model at a glance
As of 2026-11-15. Source: Anthropic models overview, individual model pages, model deprecations page and release notes.
| Model | API ID | Released | Status |
|---|---|---|---|
| Claude Opus 5.5 | claude-opus-5-5 | 2026-09-22 | Active (current Opus model) |
| Claude Opus 5 | claude-opus-5 | 2026-07-24 | Active (legacy on its model page) |
| Claude Opus 4.8 | claude-opus-4-8 | 2026-05-28 | Active (legacy on its model page) |
All three take text and image input and return text. Open weights: none. Licence: proprietary. Parameters: not stated by Anthropic.
When was each model released?
As of 2026-11-15. Dates are as shown in Anthropic's Claude Platform release notes and announcements.
| Date | Event | API ID |
|---|---|---|
| 2026-05-28 | Claude Opus 4.8 released; fast mode available as a research preview on the Claude API | claude-opus-4-8 |
| 2026-07-15 | Mid-conversation system messages available on Opus 4.8 | claude-opus-4-8 |
| 2026-07-24 | Claude Opus 5 released at $5/$25, the same pricing as Opus 4.8; mid-conversation tool changes enter beta on both | claude-opus-5 |
| 2026-08-01 | Dreams supports Opus 5 | claude-opus-5 |
| 2026-08-19 | Computer use and browser use toolsets available for Opus 5 and Opus 4.8 on the Claude API | claude-opus-5, claude-opus-4-8 |
| 2026-09-01 | Per-message effort changes in beta on Opus 5 (Claude API) | claude-opus-5 |
| 2026-09-22 | Claude Opus 5.5 released at $4/$20; fast mode research preview on the Claude API | claude-opus-5-5 |
| 2026-10-01 | Dreams supports Opus 5.5 | claude-opus-5-5 |
| 2026-10-15 | Fast mode out of research preview for Opus 5.5; also on Microsoft Foundry (Azure-hosted deployments) | claude-opus-5-5 |
| 2026-10-22 | Fast mode for Opus 5.5 on Claude Models on Google Cloud (Anthropic managed) | claude-opus-5-5 |
| 2026-10-30 | Fast mode for Opus 4.8 deprecated, removal 2026-11-30 | claude-opus-4-8 |
How much does each model cost?
As of 2026-11-15. USD per million tokens, from Anthropic's pricing page.
| Model | Input | Cached input | Cache write | Output | Batch (input / output) | Notes stated by provider |
|---|---|---|---|---|---|---|
| Claude Opus 5.5 | $4.00 | $0.20 | $5.00 (5m), $8.00 (1h) | $20.00 | $2.00 / $10.00 | Cache read 0.05x base input; fast mode $8 / $40 |
| Claude Opus 5 | $5.00 | $0.50 | $6.25 (5m), $10.00 (1h) | $25.00 | $2.50 / $12.50 | Cache read 0.1x; fast mode $10 / $50 (research preview) |
| Claude Opus 4.8 | $5.00 | $0.50 | $6.25 (5m), $10.00 (1h) | $25.00 | $2.50 / $12.50 | Cache read 0.1x; fast mode $10 / $50 (deprecated) |
Tokeniser: Anthropic's pricing page states that Claude 4.7 and later models use a newer tokeniser that "produces approximately 30% more tokens for the same text". All three Opus models here are on that tokeniser, so, unlike comparisons that cross the 4.6 to 4.7 boundary, the same prompt should produce comparable token counts on each. Confirm with your own data.
Long context and residency: The pricing page states that Claude 4.6 and later models (except Haiku 5.5) include the full 1M window at standard pricing, so there is no long-context surcharge on any of the three. Setting inference_geo to "us" applies a 1.1x multiplier to all token categories. For cost guardrails, see How to set spend limits on LLM API keys and Model Pricing Explained.
What changed between models?
Only changes Anthropic documents in its release notes and announcements are listed.
- Opus 4.7 to 4.8: Anthropic calls Opus 4.8 "a modest but tangible improvement on its predecessor" at the same price. The release notes state that with adaptive thinking enabled it triggers reasoning only when a turn needs it, and that non-default temperature, top_p or top_k return a 400 error, as on Opus 4.7. The announcement adds an effort control in Claude.ai and Cowork with a default of high, system entries inside the messages array, and fast mode at 2.5 times the speed.
- Opus 4.8 to 5: Opus 5 launched at the same $5/$25 price. The release notes record one breaking change: disabling thinking is allowed only at effort high or below, and thinking disabled with effort xhigh or max returns a 400 error. The announcement describes the model as coming "close to the frontier intelligence of Claude Fable 5 at half the price", makes it the default model on Claude Max, and introduces cyber classifiers that fall back to Opus 4.8 for flagged requests in Claude.ai, Claude Code and Claude Cowork.
- Opus 5 to 5.5: Prices drop to $4/$20, with cache reads from $0.50 to $0.20. Anthropic states about 40% lower cost on typical workloads at default settings and output more than 30% faster. Thinking can no longer be disabled: both thinking disabled and manual thinking enabled return a 400 error, and depth is set with the effort parameter. Forced tool use (tool_choice any and tool) returns a 400 error; Anthropic's advice is auto with strict tool use. Default effort moves from high to medium. Most cybersecurity tasks flagged by safeguards are re-routed to Opus 4.8.
For background on adaptive thinking and effort, see Reasoning Models Explained. Anthropic publishes an Opus 4.8 to 5.5 migration guide, linked from the Opus 4.8 model page.
What workloads does Anthropic position each model for?
This section records the provider's own positioning wording; it is not an assessment.
| Model | Positioning stated by provider | Related notes stated by provider |
|---|---|---|
| Claude Opus 5.5 | "For long-running agentic coding and knowledge work"; performs at the level of Fable 5.1 on most work | Higher five-hour usage limits on Pro, Max, Team and seat-based Enterprise plans |
| Claude Opus 5 | "A thoughtful and proactive model" close to Fable 5 "at half the price" | Default on Claude Max at launch; better than Opus 4.8 on every life sciences evaluation Anthropic ran |
| Claude Opus 4.8 | "A more effective collaborator" with improvements across benchmarks | More likely to flag uncertainty; fallback model for flagged cyber requests on Opus 5 and 5.5 |
Worked cost example
Using the list prices above, 10,000 requests a day, each with a 20K-token prompt and a 1K-token response, use 200 million input and 10 million output tokens a day. On Opus 5 or Opus 4.8 that is $1,000 input plus $250 output, $1,250 a day. On Opus 5.5 it is $800 plus $200, $1,000 a day, 20% lower. Because all three share a tokeniser, no adjustment is needed for token count. Caching widens the gap: if 15K tokens of each prompt are cache hits, Opus 5.5 charges $30 a day for those reads against $75 on Opus 5 or 4.8. Batch halves each figure. Re-run the arithmetic on measured token counts, since default effort also differs (medium on Opus 5.5, high on the other two), which changes output and thinking tokens per task.
What is the context window of each model?
As of 2026-11-15. From Anthropic's models overview and individual model pages; reasoning controls from the release notes.
| Model | Context window | Max output | Knowledge cutoff | Input | Output | Reasoning controls |
|---|---|---|---|---|---|---|
| Claude Opus 5.5 | 1M tokens | 128K tokens | Jun 2026 (reliable and training data) | Text, image | Text | Adaptive thinking, always on; default effort medium |
| Claude Opus 5 | 1M tokens | 128K tokens (300K on Batch API, beta) | May 2026 (reliable and training data) | Text, image | Text | Adaptive; default effort high; thinking can be disabled only at effort high or below |
| Claude Opus 4.8 | 1M tokens | 128K tokens (300K on Batch API, beta) | Jan 2026 (reliable and training data) | Text, image | Text | Adaptive; default effort high; non-default sampling parameters return a 400 error |
The 300K batch output option uses the output-300k-2026-03-24 beta header, as stated on the Opus 5 and Opus 4.8 model pages; the Opus 5.5 model page checked does not state a batch output limit. The knowledge cutoff moves five months between Opus 4.8 and Opus 5.5, which matters for questions about recent APIs, libraries and events.
Capabilities and tools compared
As of 2026-11-15. Anthropic's models overview states that all current models support "text and image input, text output, multilingual capabilities, vision, and tool use". Cloud IDs are from each model's overview page.
| Model | Function calling and structured outputs | Computer use tool | Fine-tuning | Cloud IDs stated by provider |
|---|---|---|---|---|
| Claude Opus 5.5 | Tool use supported; tool_choice any and tool return a 400 error; use auto with strict tool use | Anthropic positions it for computer use in its announcement; OSWorld 2.1 figure reported | Not stated by Anthropic on the pages checked (2026-11-15) | Amazon Bedrock anthropic.claude-opus-5-5; Google Cloud, Microsoft Foundry and Claude Platform on AWS claude-opus-5-5 |
| Claude Opus 5 | Tool use supported; mid-conversation tool changes in beta; automatic fallbacks in beta on the API | Computer use and browser use toolsets on the Claude API from 2026-08-19 | Not stated by Anthropic on the pages checked (2026-11-15) | Amazon Bedrock anthropic.claude-opus-5; Google Cloud, Microsoft Foundry and Claude Platform on AWS claude-opus-5 |
| Claude Opus 4.8 | Tool use supported; system entries accepted inside the messages array | Computer use and browser use toolsets on the Claude API from 2026-08-19 | Not stated by Anthropic on the pages checked (2026-11-15) | Amazon Bedrock anthropic.claude-opus-4-8; Google Cloud, Microsoft Foundry and Claude Platform on AWS claude-opus-4-8 |
For agent design questions beyond the model itself, see Computer-Use and Browser Agents.
Where is each model available?
| Model | Availability stated by provider |
|---|---|
| Claude Opus 5.5 | "All platforms", including Amazon Web Services, Google Cloud and Microsoft Azure; fast mode on the Claude API, Microsoft Foundry and Google Cloud (Anthropic managed) |
| Claude Opus 5 | All platforms at launch; default model on Claude Max; fast mode in research preview |
| Claude Opus 4.8 | Available everywhere at launch via the Claude API; fast mode removal on 2026-11-30 |
Anthropic's deprecations page states that Amazon Bedrock and Claude Models on Google Cloud (Google managed) set their own schedules. Confirm in each cloud's model catalogue.
Model cards and system cards
| Model | Model card or page | System card | Technical report |
|---|---|---|---|
| Claude Opus 5.5 | Announcement | System card | Not stated by Anthropic on the pages checked (2026-11-15) |
| Claude Opus 5 | Announcement | System card | Not stated by Anthropic on the pages checked (2026-11-15) |
| Claude Opus 4.8 | Announcement | System card | Not stated by Anthropic on the pages checked (2026-11-15) |
System card links are as given in each announcement; this page does not summarise their contents.
Benchmarks
As of 2026-11-15. Figures are provider-reported, as stated in the text of Anthropic's announcements. Opus 5.5 runs at max effort unless noted; Anthropic notes its production safeguards may have lowered Opus 5.5 scores.
| Model | Provider-reported benchmarks | Independent leaderboard results |
|---|---|---|
| Claude Opus 5.5 | Terminal-Bench 4.0 66.4% (xhigh); FrontierCode v1.1 (Main) 54.4%; CursorBench 4.0 57.8%; GDPval-AA v2.1 1846 Elo; Humanity's Last Exam (with tools) 67.7%; OSWorld 2.1 (partial) 81.8% (announcement, 2026-09-22) | Not checked for this page |
| Claude Opus 5 | In the Opus 5.5 announcement comparison: Terminal-Bench 4.0 52.3%; FrontierCode v1.1 (Main) 48.0%; CursorBench 4.0 46.6%; GDPval-AA v2.1 1708 Elo; Humanity's Last Exam (with tools) 63.6%; OSWorld 2.1 (partial) 74.0%. Its own announcement gives relative claims only | Not checked for this page |
| Claude Opus 4.8 | Not stated in the announcement text; Anthropic states the full comparison table is in the system card | Not checked for this page |
Anthropic also states that Opus 5.5 at default effort exceeds Opus 5 at max effort on Terminal-Bench 4.0 for about a fifth of the cost, and that benchmark margins are a less reliable guide at this capability level. Run your own evaluation before switching.
Deprecations and retirements
As of 2026-11-15. Source: Anthropic model deprecations page and release notes. Anthropic states it gives at least 60 days' notice before retiring publicly released models.
| Model | API ID | Notice or deprecation date | Retirement or shutdown date | Stated replacement |
|---|---|---|---|---|
| Claude Opus 5.5 | claude-opus-5-5 | Active | Not sooner than 2027-09-22 | None stated |
| Claude Opus 5 | claude-opus-5 | Active | Not sooner than 2027-07-24 | Migration to Opus 5.5 recommended on the model page |
| Claude Opus 4.8 | claude-opus-4-8 | Active | Not sooner than 2027-05-28 | Migration to Opus 5.5 recommended on the model page |
| Opus 4.8 fast mode | claude-opus-4-8 with speed "fast" | 2026-10-30 | 2026-11-30 | Claude Opus 5.5 |
After 30 November 2026, requests to claude-opus-4-8 with speed "fast" return an error. For lifecycle planning, see the AI Model Retirement Guide and the LLM Deprecation and Retirement Tracker.
Data handling and security
Anthropic's zero data retention page designates only the Fable and Mythos 5.x models as Covered Models requiring 30-day retention; none of the three Opus models is on that list, and the Opus 5.5 announcement states it is available with zero data retention "as with previous Opus models". The page states flagged content may be retained for up to 2 years under any arrangement. The Opus 5.5 announcement adds watermarking for EU AI Act compliance and "preserved thinking" for API accounts created on or after 31 August 2026, which ties thinking to the originating account; see the AI Model Provenance and Watermarking Guide. Opus 5 and 5.5 run cyber safeguards that can route requests to Opus 4.8, which matters for security teams; the Cyber Verification Program offers tiered access.
Across all three, the operational risk is the credential: a Claude API key reaches every model the workspace can call. See the Anthropic API Keys Security Guide and the LLMjacking Guide.
How to choose a model: factual criteria
| Criterion | What the documentation shows |
|---|---|
| Lifecycle runway | Opus 5.5 has the latest "not sooner than" date (2027-09-22); Opus 4.8 the earliest (2027-05-28) |
| List price per token | $4/$20 for Opus 5.5; $5/$25 for Opus 5 and 4.8 |
| Cache-heavy workloads | Cache read $0.20 on Opus 5.5 against $0.50 on Opus 5 and 4.8 |
| Ability to switch thinking off | Opus 5 at effort high or below only; not possible on Opus 5.5 |
| Forced tool use | Opus 5.5 returns a 400 error for tool_choice any and tool |
| Knowledge cutoff | Jun 2026, May 2026 and Jan 2026 respectively |
| Fast mode | Generally available on Opus 5.5; research preview on Opus 5; ends 2026-11-30 on Opus 4.8 |
For a wider view across providers, see Frontier AI Models Compared and Claude Opus vs Sonnet vs Haiku 5.5.
Practitioner checklist
- Find every workload calling claude-opus-4-8 with speed "fast" and move it before 30 November 2026.
- Pin production code to an explicit Opus API ID and record it in the AI inventory.
- Before moving to Opus 5.5, remove thinking disabled and manual thinking parameters and replace forced tool_choice with auto plus strict tool use.
- Re-baseline cost per task, since default effort changes from high to medium on Opus 5.5.
- Test security-related prompts for safeguard fallbacks to Opus 4.8 and record which model actually answered.
- Check Bedrock and Google-managed Vertex catalogues separately for their own retirement dates.
- Confirm data retention terms against your agreement and store Claude API keys in a secrets manager with spend limits.
Standards and references
- Anthropic: Model deprecations (accessed 2026-11-15)
- Anthropic: Models overview (accessed 2026-11-15)
- Anthropic: Pricing (accessed 2026-11-15)
- Anthropic: Claude Platform release notes (accessed 2026-11-15)
- Anthropic: Claude Opus 5.5 model overview (accessed 2026-11-15)
- Anthropic: Claude Opus 5 model overview (accessed 2026-11-15)
- Anthropic: Claude Opus 4.8 model overview (accessed 2026-11-15)
- Anthropic: Zero data retention (accessed 2026-11-15)
- Anthropic: Introducing Claude Opus 5.5 (accessed 2026-11-15)
- Anthropic: Introducing Claude Opus 5 (accessed 2026-11-15)
- Anthropic: Introducing Claude Opus 4.8 (accessed 2026-11-15)
Related NHI Mgmt Group resources: Anthropic Models, Claude Opus 5.5, Claude Sonnet Versions Compared, LLM Pricing Tracker, Claude for Enterprise Security Guide.