Last updated: October 13, 2026 · Model IDs, release date, lifecycle status, availability and pricing checked against Anthropic’s official model pages, release notes, pricing page and model deprecations table.
Claude Opus 4.8 model ID: claude-opus-4-8 on the Claude API, Google Cloud (Vertex AI), Microsoft Foundry and Claude Platform on AWS; anthropic.claude-opus-4-8 on Amazon Bedrock. Release date: May 28, 2026. Status as of October 13, 2026: still available, but Anthropic now labels it a legacy model. It was superseded by Claude Opus 5 (July 24, 2026) and then by Claude Opus 5.5 (claude-opus-5-5, September 22, 2026), the current Opus model. Retirement is “not sooner than May 28, 2027”. Price: $5 / $25 per million input / output tokens, 1M-token context, 128K max output.
Claude Opus 4.8 model ID on every platform
The exact model name you send in the model field depends on where you call Claude. These are the identifiers Anthropic lists on its official Claude Opus 4.8 model page, checked on October 13, 2026:
| Platform | Opus 4.8 model ID | Notes |
|---|---|---|
| Claude API (Anthropic) | claude-opus-4-8 | Dateless ID, but a pinned snapshot; it never moves to a newer model |
| Amazon Bedrock | anthropic.claude-opus-4-8 | Bedrock Messages-API endpoint; Bedrock sets its own lifecycle dates |
| Google Cloud (Vertex AI) | claude-opus-4-8 | Global, multi-region and regional endpoints; Google sets its own lifecycle dates |
| Microsoft Foundry | claude-opus-4-8 | Default deployment name; you send the deployment name you chose |
| Claude Platform on AWS | claude-opus-4-8 | Uses Claude API IDs, not Bedrock-style IDs |
There is no separate “alias” for Opus 4.8. From the 4.6 generation on, Anthropic’s dateless IDs are themselves pinned snapshots, so claude-opus-4-8 will keep returning Opus 4.8 until the model is retired. Anthropic’s Models API returns the same ID together with live token limits if you want to check it programmatically.
As of October 13, 2026, Anthropic’s models overview lists Claude Opus 4.8 under “Legacy models (still available)”. The current Opus model is Claude Opus 5.5 (claude-opus-5-5), released September 22, 2026 at $4 / $20 per million tokens, which is cheaper than Opus 4.8. Anthropic’s own Opus 4.8 page says: “consider migrating to Claude Opus 5.5.” If you only came for the newest model name, use claude-opus-5-5.
When was Claude Opus 4.8 released? Release date and timeline
Claude Opus 4.8 was released on May 28, 2026. Anthropic’s announcement (“Introducing Claude Opus 4.8”) and the Claude Platform release notes for that day both date it, and it was available immediately on the Claude API at the same price as Opus 4.7. Here is how the Opus line has moved since then:
| Date | Model | Model ID | What it meant for Opus 4.8 |
|---|---|---|---|
| Apr 16, 2026 | Claude Opus 4.7 | claude-opus-4-7 | The predecessor Opus 4.8 replaced |
| May 28, 2026 | Claude Opus 4.8 | claude-opus-4-8 | Launched as Anthropic’s most capable generally available model |
| Jul 24, 2026 | Claude Opus 5 | claude-opus-5 | Described by Anthropic as “a step-change improvement over Claude Opus 4.8”, same $5 / $25 price |
| Sep 22, 2026 | Claude Opus 5.5 | claude-opus-5-5 | Current Opus model; Opus 4.8 is now listed as legacy |
| Not sooner than May 28, 2027 | Earliest possible Opus 4.8 retirement | Anthropic’s commitment for the platforms it operates; Bedrock and Google Cloud set their own dates | |
Is Claude Opus 4.8 still available? Status and availability (October 2026)
Yes, but as a legacy model. Anthropic’s model deprecations table lists claude-opus-4-8 as Active, with no deprecation date and a retirement date of “not sooner than May 28, 2027”. In Anthropic’s terms, legacy means the model “will no longer receive updates and may be deprecated in the future”. Anthropic also commits to at least 60 days’ notice before it retires a publicly released model.
| Status field | Verified answer (Oct 13, 2026) |
|---|---|
| Released | May 28, 2026 |
| Lifecycle status | Active (legacy); successor: Claude Opus 5.5 |
| Retirement | Not sooner than May 28, 2027 |
| API platforms | Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS |
| Price | $5 input / $25 output per million tokens (Batch: $2.50 / $12.50) |
| Context / max output | 1M tokens / 128K tokens (300K on the Batch API with a beta header) |
| Knowledge cutoff | January 2026 (reliable knowledge and training data) |
| Fast mode | Research preview on the Claude API only, $10 / $50 per million tokens |
Claude apps and Claude Code
At launch, Anthropic said it was “upgrading Claude Opus to a new version” and that Opus 4.8 was “available today”. Since then the defaults have moved on. Anthropic’s Opus page now says Opus 5.5 is the Opus model available to Pro, Max, Team and Enterprise users. Its help docs do not say whether Opus 4.8 is still in the claude.ai model picker, so check the picker on your own plan rather than relying on third-party lists.
In Claude Code, the opus alias now points to Opus 5.5 on the Anthropic API, Amazon Bedrock and Google Cloud. To pin Opus 4.8, use the full model name, for example /model claude-opus-4-8, claude --model claude-opus-4-8 or ANTHROPIC_MODEL=claude-opus-4-8. One detail many teams miss: Claude Code still uses Opus 4.8 behind the scenes as the fallback model for cybersecurity-flagged requests that Opus 5.5, Opus 5 and the Fable models refuse. On Bedrock, Google Cloud and Foundry, that fallback only works if an Opus fallback target resolves in your deployment. You get one by keeping an Opus 4.8 entry in the provider’s model list or by setting ANTHROPIC_DEFAULT_OPUS_MODEL. That is a practical reason not to remove Opus 4.8 from those deployments yet.
Opus 4.8 vs Opus 5 vs Opus 5.5: which model name to use now
| Claude Opus 4.8 | Claude Opus 5 | Claude Opus 5.5 | |
|---|---|---|---|
| Claude API ID | claude-opus-4-8 | claude-opus-5 | claude-opus-5-5 |
| Released | May 28, 2026 | Jul 24, 2026 | Sep 22, 2026 |
| Status (Oct 13, 2026) | Legacy | Legacy | Current (latest) |
| Input / output per MTok | $5 / $25 | $5 / $25 | $4 / $20 |
| Thinking | Adaptive, off unless requested | On by default | Adaptive, always on |
| Default effort (API) | high | high | medium |
| Context / max output | 1M / 128K | 1M / 128K | 1M / 128K |
| Retirement | Not before May 28, 2027 | Not before Jul 24, 2027 | Not before Sep 22, 2027 |
For a new project, Anthropic’s docs recommend starting with Opus 5.5. Keep claude-opus-4-8 when you have a validated production workload you do not want to re-baseline yet, when you rely on running without thinking, when you use forced tool choice (tool_choice any/tool, which returns a 400 error on Opus 5.5), or when you have a Priority Tier commitment (Opus 5.5 does not support Priority Tier). For the rest of the 2026 lineup, see our Claude Sonnet 5.5 pricing and migration guide and the Claude Fable pricing breakdown.
What changed in Claude Opus 4.8 (at launch)
This section describes Opus 4.8 as it shipped on May 28, 2026. Opus 4.8 built on Opus 4.7 rather than introducing a new API family. Anthropic highlighted better long-horizon agentic coding, tool triggering, compaction recovery and effort calibration, plus several concrete platform changes.
Dynamic workflows are now generally available
Dynamic workflows let Claude Code plan a large task, dispatch tens to hundreds of parallel subagents, verify their work and return one coordinated result. The feature launched as a research preview with Opus 4.8, but Anthropic’s current product page now says it is generally available in Claude Code CLI, Desktop and the VS Code extension for Pro, Max, Team and Enterprise plans, as well as on supported API and cloud surfaces.
This is useful for decomposable work such as codebase-wide audits, migrations and independent verification. It is not a free performance multiplier: Anthropic warns that workflows can consume substantially more tokens than a typical Claude Code session. Start with a scoped task and measure usage before making workflows a default.
For a broader explanation of the orchestration pattern, see our agentic workflows guide.
Mid-conversation system messages
Opus 4.8 accepts a role: "system" message immediately after a user turn in the Messages API, subject to Anthropic’s placement rules. This lets an application update instructions during a long-running conversation without rebuilding the full history, which can preserve prompt-cache hits. The top-level system field remains the correct place for instructions that apply from the beginning.
Effort defaults and adaptive thinking
The effort parameter defaults to high across Opus 4.8 surfaces. Adaptive thinking is supported, but API requests without a thinking field run without thinking. Enable it with {"type":"adaptive"} and set depth through output_config.effort. Manual extended-thinking budgets are not supported on Opus 4.8.
Fast mode remains a research preview
Fast mode uses the same Opus 4.8 model with up to 2.5× higher output tokens per second. It is not a different or more capable model, and the benefit targets output speed rather than time to first token. As of October 13, 2026 it is still a research preview on the Claude API only (including Claude Managed Agents), and not available on Amazon Bedrock, Google Cloud or other cloud platforms. Fast mode now also supports Opus 5 and Opus 5.5.
Lower cache minimum and documented refusal details
The minimum cacheable prompt length is 1,024 tokens, down from 2,048 on Opus 4.7. Anthropic also documents the stop_details object on refusals, which helps applications route different refusal categories without treating them as generic failures.
Claude Opus 4.8 pricing, context window and Fast mode
| Mode | Input / MTok | Output / MTok | Status and use |
|---|---|---|---|
| Standard Opus 4.8 | $5 | $25 | Standard Claude API pricing (unchanged as of Oct 13, 2026) |
| Fast mode | $10 | $50 | Research preview; latency-sensitive output streaming |
| Batch API | $2.50 | $12.50 | Asynchronous workloads; Fast mode is not available with Batch |
| Prompt caching | $6.25 write (5 min) · $10 write (1 h) · $0.50 read | n/a | Minimum cacheable prompt: 1,024 tokens |
| Opus 5.5, for comparison | $4 | $20 | Current Opus model; Fast mode $8 / $40 |
The official model overview lists a 1M-token context window and 128K maximum output for Opus 4.8 on the synchronous Messages API. The 1M window is included at standard per-token pricing. Provider-specific endpoint limits and identifiers can differ, so check the current platform documentation before assuming identical behavior across every cloud.
Fast mode is exactly 2× the standard input and output price. Because it accelerates output token generation rather than model intelligence, use it only where response speed has measurable product value. Requests at fast and standard speeds do not share prompt-cache prefixes, and a fast-mode rate or capacity error does not silently fall back to standard speed.
Claude Opus 4.8 benchmarks: what the results do and do not show
Anthropic’s launch materials report improvements across coding, agentic work, reasoning and practical knowledge tasks. These results are useful release signals, but most are vendor-run or partner-reported. They should guide what to test in your own harness, not replace workload-specific evaluation.
| Reported result | What it indicates | Important limit |
|---|---|---|
| 84% on Online-Mind2Web | Strong browser-agent and computer-use signal | Partner-reported in Anthropic’s announcement |
| About 4× less likely to let self-written code flaws pass unremarked | Better uncertainty and flaw flagging | Not a claim that the model writes 4× fewer bugs |
| First reported model above 10% all-pass on Legal Agent Benchmark | Progress on long multi-step professional workflows | All-pass is strict and the result is partner-reported |
| Improved long-horizon coding, tool triggering and compaction recovery | More reliable agentic work than Opus 4.7 | Re-test against your own prompts, tools and cost limits |
Anthropic says Opus 4.8 is around four times less likely than its predecessor to allow flaws in code it wrote to pass unremarked. That is an honesty/self-review measure, not a blanket fourfold improvement in code correctness.
Migrating to and from Claude Opus 4.8
From Opus 4.7 to Opus 4.8
Anthropic documented no breaking API changes for code already running on Opus 4.7. The required change is the model name. The real work is re-baselining behavior, latency and cost at your chosen effort level, because Opus 4.8 defaults to high effort and only thinks when you enable adaptive thinking.
from anthropic import Anthropic
client = Anthropic()
response = client.messages.create(
model="claude-opus-4-8", # pinned Opus 4.8 snapshot (was claude-opus-4-7)
max_tokens=16000,
thinking={"type": "adaptive"}, # omit to run without thinking on Opus 4.8
output_config={"effort": "high"},
messages=[
{"role": "user", "content": "Review this migration plan."}
],
)
- Swap the model ID from
claude-opus-4-7toclaude-opus-4-8(on Bedrock:anthropic.claude-opus-4-8). - Re-run your evals and re-baseline latency and cost; set
effortexplicitly if your product depends on a stable trade-off. - Remove obsolete thinking configuration. Use adaptive thinking and
output_config.effort, notbudget_tokens. Non-defaulttemperature,top_portop_kvalues return a 400 error, as on Opus 4.7. - Remove an old long-context beta header if your client still sends one. The 1M window is the default on supported endpoints.
From Opus 4.8 to Opus 5.5 (the current Opus)
Moving off Opus 4.8 is no longer just a model-name swap. Anthropic’s Opus 5.5 migration guide lists these changes for code written against claude-opus-4-8:
- Thinking is always on. On Opus 4.8, a request without a
thinkingfield runs without thinking. On Opus 5.5 the same request thinks, andthinking: {"type": "disabled"}returns a 400 error. Thinking tokens are billed as output, so revisitmax_tokensand lowereffortwhere you want less thinking. - Lower default effort. Opus 5.5 defaults to
mediuminstead ofhigh, so set effort explicitly if you need to match Opus 4.8 behavior. - No forced tool use.
tool_choicetypesanyandtoolreturn a 400 error; useautowith strict tool use or structured outputs. - Computer use toolset. On the Claude API and Google Cloud, Opus 5.5 needs
computer_toolset_20260801instead ofcomputer_20251124. - No Priority Tier on Opus 5.5, while Opus 4.8 keeps it.
- Cheaper and easier to cache. $4 / $20 instead of $5 / $25, and the minimum cacheable prompt drops from 1,024 to 512 tokens.
For a practical coding-tool comparison, see Claude Code vs Cursor and our Claude Code pricing guide. For the earlier generation, see our Claude Opus 4.7 explainer.
FAQ
What is the Claude Opus 4.8 model ID?
The Claude API model ID is claude-opus-4-8. The same ID is used on Google Cloud (Vertex AI), Microsoft Foundry and Claude Platform on AWS. On Amazon Bedrock it is anthropic.claude-opus-4-8. The dateless ID is a pinned snapshot, so it always returns Opus 4.8.
When was Claude Opus 4.8 released?
Anthropic released Claude Opus 4.8 on May 28, 2026. It was followed by Claude Opus 5 on July 24, 2026 and Claude Opus 5.5 on September 22, 2026.
What is the Opus 4.8 model name in the API?
Send “claude-opus-4-8” as the model parameter in the Messages API. Its display name is Claude Opus 4.8. There is no separate dated snapshot name such as claude-opus-4-8-2026xxxx.
Is Claude Opus 4.8 still available?
Yes. As of October 13, 2026, Anthropic lists Claude Opus 4.8 as an active legacy model on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS. Retirement is not sooner than May 28, 2027, and Anthropic gives at least 60 days’ notice before it retires a model.
What replaced Claude Opus 4.8?
Claude Opus 5.5 (claude-opus-5-5), released September 22, 2026, is the current Opus model. It costs $4 / $20 per million input / output tokens versus $5 / $25 for Opus 4.8. Claude Opus 5 (claude-opus-5, July 24, 2026) came in between and is also legacy now.
How much does Claude Opus 4.8 cost?
$5 per million input tokens and $25 per million output tokens on the Claude API. Batch API is $2.50 / $12.50, cache reads are $0.50 per million tokens, and Fast mode (research preview, Claude API only) is $10 / $50.
What context window does Claude Opus 4.8 have?
1M tokens, with up to 128K output tokens on the synchronous Messages API and up to 300K on the Message Batches API with the output-300k-2026-03-24 beta header. Its reliable knowledge cutoff is January 2026.
Update log
- October 13, 2026: Added an answer-first box with the model ID on every platform and the release date. Updated the status to legacy following Claude Opus 5 (July 24) and Claude Opus 5.5 (September 22). Added the release timeline, lifecycle and retirement data, the Opus 4.8 vs 5 vs 5.5 comparison, the Opus 4.8 → 5.5 migration changes, and the Claude Code alias and cyber-fallback details. Rewrote the FAQ around model ID and release-date questions.
- July 18, 2026: Rebuilt the page around release status, model ID and availability. Reverified pricing, context and migration guidance, and updated dynamic workflows to generally available.
- June 2, 2026: Original release-week analysis published after the May 28 launch.
Sources and further reading (14)
Model IDs, release dates, lifecycle status, availability and prices were checked against official Anthropic and Claude documentation on October 13, 2026. Benchmark claims are identified as Anthropic- or partner-reported where applicable.
- Claude Platform Docs: Claude Opus 4.8 model page (legacy status, model IDs per platform, release date, retirement, pricing, knowledge cutoff).
- Claude Platform Docs: Models overview (current lineup and legacy list; pinned-snapshot ID rule).
- Claude Platform Docs: Model deprecations (Active status, retirement not sooner than May 28, 2027, 60-day notice).
- Claude Platform Docs: Release notes (May 28, July 24 and September 22, 2026 launches).
- Anthropic: Introducing Claude Opus 4.8 (May 28, 2026 launch and reported evaluations).
- Anthropic: Introducing Claude Opus 5 (July 24, 2026; same price as Opus 4.8).
- Anthropic: Claude Opus 5.5 (September 22, 2026 launch, pricing and availability).
- Claude Platform Docs: Opus 5.5 migration guide (changes when moving from Opus 4.8).
- Claude Platform Docs: Pricing (standard, cache, Batch and Fast mode rates).
- Claude Platform Docs: Fast mode (supported models and research-preview status).
- Claude Code Docs: Model configuration (alias resolution, pinning a model, automatic fallback to Opus 4.8).
- Anthropic: Claude Opus product page (Opus 5.5 availability on Pro, Max, Team and Enterprise).
- Claude: Introducing dynamic workflows in Claude Code (GA status and plan availability).
- Anthropic: Claude Opus 4.8 System Card (evaluation and safety methodology).

