Claude Opus 5.5 Guide (October 2026): Fast Mode GA, Pricing, Google Cloud and Benchmarks

Updated 2026-10-26: Fast mode on Opus 5.5 is now GA with no beta header and is available on Microsoft Foundry and the new Anthropic-managed Google Cloud offering. We added the Sonnet 5.5 and Haiku 5.5 launches, the API version 2026-10-21 changes, refreshed Artificial Analysis rankings and a cost table for each effort level.

Anthropic shipped Claude Opus 5.5 on 22 September 2026, about two months after Claude Opus 5. The release cut per-token prices by 20% and the cache-read price by 60%. In Anthropic's own capability summary (Table 8.1.A of the system card), Opus 5.5 scores above both Opus 5 and the more expensive Claude Fable 5.1 on every row. Since launch, the rest of the 5.5 family has shipped (Sonnet 5.5 on 28 September and Haiku 5.5 on 7 October). Fast mode has left research preview, and Anthropic now runs its own Claude API offering inside Google Cloud.

This guide is for engineers and engineering leads deciding whether to build on Opus 5.5 and how to configure it. It covers model IDs, limits, pricing against the rest of the lineup, fast mode, where to run the model, effort settings and what each one costs, benchmarks (vendor-reported and independent), what changed from Opus 5, and the safety findings that affect agent design. Unless a number is marked as third-party, it comes from Anthropic's docs, announcement or system card.

What is Claude Opus 5.5?

Anthropic's model page describes Claude Opus 5.5 as the model "for long-running agentic coding and knowledge work." In the current lineup it sits between Claude Sonnet 5.5 ($2/$10) and Claude Fable 5.1 ($10/$50) on price. It beats Fable 5.1 on every row of Anthropic's capability summary (Table 8.1.A), though Fable 5.1 still leads on OfficeQA and OfficeQA Pro elsewhere in the system card.

Three things set it apart from Claude Opus 5:

• It costs less. It is $4/$20 instead of $5/$25, and cache reads are $0.20 instead of $0.50. Anthropic's announcement says it is 40% cheaper than Opus 5 on typical workloads at default settings, and that output is more than 30% faster.
• The default effort is lower. A request that omits effort runs at medium. On Opus 5 the default was high.
• Thinking is always on. Adaptive thinking can't be disabled, and forced tool use returns an error. Both are breaking API changes, covered in our Opus 5.5 migration guide.

Claude Opus 5.5 specs and model IDs

The API ID is claude-opus-5-5. It has no date suffix, and the same string works on every platform except Amazon Bedrock.

A note on the context window: Claude models from Opus 4.7 onward use a newer tokenizer. On it, 1M tokens holds about 555k words, where older models fit about 750k. If you are porting long-document prompts from a model older than Opus 4.7, budget for that.

How much does Claude Opus 5.5 cost?

On the Claude API, Opus 5.5 costs $4 per million input tokens and $20 per million output tokens. The table below compares it with the rest of the current lineup, using Anthropic's pricing page as of October 2026 (USD per million tokens):

Several details affect real bills:

• Cache reads use a lower multiplier. Most Claude models charge 10% of the input price for a cache hit. Opus 5.5 and Sonnet 5.5 charge 5%. For agent loops that re-read a large, stable prefix every turn, this is the largest saving in the release. Our explainer on how prompt caching reduces AI cost covers how to structure prompts to get cache hits.
• There is no long-context premium. The full 1M window bills at the standard rate, so a 900k-token request costs the same per token as a 9k-token one. Haiku 5.5 is the exception in the lineup: prompts over 100,000 tokens pay $0.50/$2.50.
• US-only inference costs 10% more. Setting inference_geo: "us" applies a 1.1x multiplier to every token category, and the multiplier stacks on fast mode pricing.
• Batch is half price ($2/$10), but fast mode can't be combined with it.

Claude Opus 5.5 fast mode: now generally available

At launch, fast mode was a research preview on the first-party API only, and it required the fast-mode-2026-02-01 beta header. That changed on 15 October 2026. According to the release notes and the fast mode docs:

• No header and no access request. Set speed: "fast" on an Opus 5.5 request. Requests that still send the old beta header keep working.
• Same model, faster serving. Anthropic describes fast mode as the same model with a faster inference configuration, giving up to 2.5x higher output tokens per second with no change in intelligence. The speedup applies to output throughput, not to time to first token, so it is most visible when you stream.
• Pricing is $8/$40 per million tokens, double the standard rate, across the full context window. Prompt caching and data-residency multipliers apply on top.
• Default rate limits. Opus 5.5 fast mode has a dedicated rate limit by default in the Start, Build and Scale usage tiers. Organizations in the Evaluation tier get it once they move up. When you exceed the limit, the API returns a 429 with a retry-after header.
• Platforms. Fast mode works on the Claude API (including Claude Managed Agents), on Microsoft Foundry since 15 October (Azure-hosted deployments only, not deployments hosted on Anthropic), and on Claude Models on Google Cloud (Anthropic managed) since 22 October. The fast mode docs do not list Amazon Bedrock, Claude Platform on AWS or the Google-managed Google Cloud offering.
• Exclusions. Fast mode is not available with the Batch API or with a Priority Tier commitment.
• Older Opus models stay in preview. Fast mode on Opus 5 and Opus 4.8 is still a research preview at $10/$50. It still needs the beta header, and requests without it return a 400.

The response's usage.speed field tells you which speed served the request. The SDK retries fast-mode 429s twice by default. If you would rather drop to standard speed than wait, catch the rate-limit error and resend without speed. Doing that causes a prompt-cache miss, because requests at different speeds don't share cached prefixes.

import anthropic

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-opus-5-5",
    max_tokens=4096,
    speed="fast",  # GA on Opus 5.5: no beta header needed
    messages=[{"role": "user", "content": "Refactor this module to use dependency injection"}],
)

print(response.usage.speed)  # "fast" or "standard"
text = "".join(b.text for b in response.content if b.type == "text")
print(text)

What fast mode costs: a worked example

Take one agent turn with a 50,000-token prompt and a 4,000-token response, where 45,000 of the prompt tokens are a cache hit.

Fast mode doubles the bill for the same tokens. It is worth paying for when a person is waiting on long streamed outputs, such as interactive coding sessions or live review. Background and batch work should stay on standard speed. Lowering the effort level is usually the cheaper way to cut latency, because it reduces the number of thinking tokens the model generates.

Where is Claude Opus 5.5 available?

• Claude API: all customers.
• AWS: Amazon Bedrock (anthropic.claude-opus-5-5) and Claude Platform on AWS.
• Google Cloud: two options, covered below.
• Microsoft Foundry: available, with fast mode on Azure-hosted deployments.
• Claude apps and Claude Code: the launch announcement says five-hour usage limits went up on the Pro, Max, Team and seat-based Enterprise plans, and that subscribers get a rate limit reset they can save and use when they choose. It also says fast mode is available in Claude Code. Check /model in your install to see which Opus is your default.

Claude Opus 5.5 on Google Cloud: Anthropic managed vs Google managed

Since 22 October 2026, Google Cloud customers have two ways to run Claude. The Google-managed offering is the existing route, which Google operates and prices. The new Claude Models on Google Cloud (Anthropic managed) is the Claude API run by Anthropic and reached through your Google Cloud project. Google Cloud IAM controls access, and billing goes through Google Cloud Marketplace. The Google-managed column below draws on Anthropic's Google-managed docs.

Anthropic lists Managed Agents (beta), code execution, computer use, agent skills, the Files API, citations, web search and fetch, the MCP connector (beta), prompt caching, batch processing and the Claude Console as part of the Anthropic-managed offering. Anthropic also advises confirming existing discounts with your account team before you start, because discounts can't be applied retroactively. One API difference to note: Anthropic's what's-new page says Opus 5.5 rejects the older computer_20251124 computer-use tool on the Claude API and on Google-managed Google Cloud, where you need computer_toolset_20260801. Amazon Bedrock still accepts the older tool. The page doesn't name the Anthropic-managed offering in this rule. Because that offering runs the Claude API, plan on the toolset there too, and test before you rely on the older tool.

API version 2026-10-21: what Opus 5.5 users should check

On 21 October Anthropic released API version 2026-10-21. The new SDK majors send it by default: Python 2.0.0, TypeScript 1.0.0, Go 2.0.0, Java 3.0.0, C# 13.0.0, PHP 1.0.0, Ruby 2.0.0 and the ant CLI 2.0.0. These releases also move from the MIT license to Apache 2.0. The changes most likely to break Opus 5.5 code are:

• Sampling parameters are removed. Sending temperature, top_p or top_k returns a 400, even at the default value. Claude 4.7 and later models already accepted only the defaults, so Opus 5.5 output doesn't change, but any leftover temperature=0 in your code now fails.
• Message shape is stricter. document blocks placed directly in assistant messages, and non-empty citations on user text blocks, now return 400s.
• The SDKs authenticate with Authorization: Bearer instead of X-Api-Key (the Microsoft Foundry clients still send x-api-key). Update any proxy, gateway or log filter that reads the old header.

Our API 2026-10-21 and SDK migration guide walks through the upgrade language by language.

What effort level should you use?

Opus 5.5 has five effort levels: low, medium, high, xhigh and max. A request that omits effort runs at medium. Two changes from Opus 5 pull in opposite directions:

1. The default is lower, so an integration you don't change usually spends fewer tokens per call.
2. At a given level, the model thinks more per turn than Opus 5 did, "most of all at xhigh and max," according to the what's-new page. If you pinned max on Opus 5, the same setting on Opus 5.5 can cost more.

Anthropic's advice is to re-run your effort sweep instead of carrying a setting over, and to leave room in max_tokens for thinking. Our Opus 5 pricing and effort breakdown covers how the effort levels worked on the previous model. Independent data shows how steep the trade-off is on Opus 5.5. These are Artificial Analysis Intelligence Index results for each Opus 5.5 effort setting, with the cost to run each index task, as of October 2026:

Source: Artificial Analysis leaderboard (independent). Percentages are our arithmetic: for example, $1.82 / $5.98 = 0.304.

The step from high to max buys 4 index points for 3.3x the cost ($5.98 / $1.82 = 3.29). For most production agents, high or xhigh is the practical ceiling. Keep max for hard tasks where one extra solved problem is worth several times the spend. The drop from medium to low costs 9 points, the largest single step, so test low on your own evals before using it as a latency fix.

The system card points the same way. On CursorBench 4.0 (measured and reported by Cursor), Opus 5.5 scored 57.8% at max, 56.0% at both xhigh and high (about $4 per task at high), and 52.5% at medium for about $3 per task. The Opus 5.5 costs are Anthropic's estimates from Cursor's token counts at list prices. Even the medium score beats Fable 5.1 at max effort (51.8% at $17.28 per task). On FrontierCode, Opus 5.5's best scores (54.6% on Main, 65.3% on Extended) came at medium effort. In the announcement, coding-agent company Factory calls it the "first model we'd default to at medium effort."

A minimal Python call with an explicit effort level:

import anthropic

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-opus-5-5",
    max_tokens=32000,                   # leave headroom for thinking
    output_config={"effort": "high"},   # low | medium | high | xhigh | max
    messages=[{"role": "user", "content": "Review this diff for concurrency bugs: ..."}],
)

# Responses can open with thinking blocks, so select by type, not position
text = "".join(b.text for b in response.content if b.type == "text")
print(text)

Don't send thinking: {"type": "disabled"} or a manual budget_tokens. Both return a 400 on Opus 5.5. For lower latency, use a lower effort level, fast mode, or Sonnet 5.5 or Haiku 5.5.

Claude Opus 5.5 benchmarks

The table below is Anthropic's capability summary (Table 8.1.A, system card p.174). GPT-5.6 Sol, CursorBench 4.0 and Chartography figures are added from the announcement and system card §8.8 and §8.13.1. Every figure comes from Anthropic's publications rather than our own testing. For the Table 8.1.A rows, Claude results use adaptive thinking at max effort averaged over five trials, except Terminal-Bench 4.0, which is reported at xhigh. CursorBench was measured and reported by Cursor in its own harness. Competitor figures come from developers' published results or benchmark leaderboards, not from Anthropic re-running them. A dash means no number was published. The announcement labels the computer-use benchmark OSWorld 2.1; we follow the system card, which calls it OSWorld 2.0.

Sources: Claude Opus 5.5 system card, §8, and the announcement. Competitor figures are from OpenAI's published results or the relevant benchmark leaderboards (Artificial Analysis for GDPval-AA and AA-Briefcase, Zapier for AutomationBench, Cursor for CursorBench). The announcement notes that Opus 5.5 was evaluated with production safeguards on, and that AutomationBench was run without fallback models, so safeguard interventions counted as failures.

What the table shows:

• Coding is the headline. SWE-bench Pro is up 10.7 points over Opus 5, and Terminal-Bench 4.0 is up 14.1. Treat the Terminal-Bench figure as a vendor self-run. It was measured at xhigh, with a standard error of ±2.6 points. In Artificial Analysis's independent Terminal-Bench 4.0 run, listed on its Opus 5.5 model page, Opus 5.5 (max with fallback) scored 59.6%. That is 6.8 points below Anthropic's self-reported 66.4.
• Vision improved the most. On Chartography without tools, Opus 5.5 scored 64.4%, against 44.8% for Fable 5.1 and 29.8% for Opus 5. That matters if you feed the model dashboards, screenshots or diagrams.
• GPT-6 Astra still wins some rows. It leads on Terminal-Bench-Science (64.6 vs 58.7) and AutomationBench (41.4 vs 40.0). On FrontierSWE v2 (run by Proximal), GPT-6 Astra scored 65.5% to Opus 5.5's 62.3%. See our GPT-6 Astra guide for that model. Our Opus 5.5 vs GPT-6 Astra vs Grok 4.7 comparison covers these head to head.
• Fable 5.1 still leads in places. It stays ahead on OfficeQA (80.2% vs 78.9%) and OfficeQA Pro (69.0% vs 67.7%). Anthropic also writes that, in its own use, "the gap between Opus 5.5 and Claude Fable 5.1 is narrower than these scores suggest."

Other results from §8 of the system card: 74.2% on DeepSWE v1.1; 91.2% on ArXivMath (August 2026) without tools and 96.9% with tools (Fable 5.1: 82.9% and 92.1%); 91.2% on the long-context ProgramBench (Fable 5.1: 87.6%, Opus 5: 85.4%); 77.8% pass@1 on Toolathlon Verified; and 94.3% average accuracy on GMMLU across 42 languages. For how to read these benchmark families, see our Opus 5 benchmarks explainer.

Anthropic did not publish SWE-bench Verified, GPQA Diamond or ARC-AGI results for Opus 5.5. Its SWE-bench reporting now uses Pro, Multilingual and Multimodal, so any Opus 5.5 SWE-bench Verified score you see is third-party.

How does Opus 5.5 rank on the Artificial Analysis Intelligence Index?

It ranks first. Independent benchmarker Artificial Analysis lists Claude Opus 5.5 at 58 on its Intelligence Index v4.3.2 (max effort, with fallback). That puts it #1 on AA's leaderboard as of 26 October 2026, ahead of the top entry from every other lab. The highest non-Anthropic entry is Meta's Muse Wonder Full (max) at 56, for which AA lists no cost per task. AA scores each effort setting separately, so name the variant whenever you quote one of these figures.

Source: Artificial Analysis leaderboard, accessed 26 October 2026. The first ten rows are the top ten entries in leaderboard order. *GPT-6.1 Sol (max) is one selected entry from the 52-point tier, which also includes Muse Wonder (max), GPT-6 Astra (xhigh) and Claude Sonnet 5.5 (xhigh). Scores are rounded as AA displays them.

The cost column matters for budgeting. At max effort, AA reports that Opus 5.5 used 260M output tokens to run the full index, against 140M for Opus 5 (max) on AA's Opus 5 page. So despite the 20% per-token price cut, AA measured $5.98 per index task for Opus 5.5 (max) against $5.86 for Opus 5 (max). Per finished job at max effort, it is not cheaper. The per-token saving shows up at xhigh and below.

Max effort is also not the cheapest route to a high score. GPT-6.1 Sol (max) reaches 52 for $0.72 per task and Gemini 4 Argon (high) reaches 53 for $1.62, against $5.98 for Opus 5.5 (max). Below max the picture flips. Opus 5.5 (high) reaches 54 for $1.82, less than GPT-6 Astra (max) costs for 53 ($3.26). Kimi K3.1 Pro reaches 55 for $2.11, close to Opus 5.5 (xhigh) at 56 for $3.46. Opus 5.5 is cost-efficient below max effort, not at it. At high it beats Fable 5.1 at max effort for under a quarter of the cost ($1.82 / $7.63 = 0.24). Sonnet 5.5 at max ties Opus 5.5 at xhigh on AA's rounded score (56) but costs more per task ($5.46 vs $3.46), which is one reason to compare models at the effort level you will actually run. Our Sonnet 5.5 vs Opus 5.5 comparison goes deeper, and our Opus 5.5 vs GPT-6 Sol comparison covers the OpenAI side of the cost curve.

AA runs its own harnesses, so its numbers won't match Anthropic's self-reported benchmark figures. Neither is wrong. Say which one you are quoting.

What's new in Claude Opus 5.5 vs Opus 5?

Anthropic says Opus 5.5 reaches the same quality in fewer tokens and steps, and backs this with customer results in the announcement:

• Optiver says it matched Opus 5's quality "in about half the turns, time and output tokens."
• Kiro reports about 40% fewer calls and half the tokens on a public command-line benchmark.
• Deloitte says Opus 5.5 caught 72% of known bugs at its lowest effort setting, against 56% for Opus 5 at high.
• Hebbia reports 86.6% coverage on finance workflows, against 60.3% for Opus 5.

These are vendor-selected customer results, not independent tests.

One change fails silently: progress text. If your app streams the short notes the model writes between tool calls, those notes now arrive as thinking blocks, and the blocks are empty at the default display: "omitted". Your UI goes quiet with no error. Thinking blocks are also bound to the conversation. For accounts created on or after 31 August 2026, editing the system prompt, tools or earlier messages and then replaying thinking returns a 400, so keep conversations append-only. The Opus 5.5 migration guide covers these changes and each breaking change with before-and-after code.

What does the Opus 5.5 system card say about safety?

The system card (230 pages, dated 22 September 2026) includes several findings that matter for builders:

• Chemical and biological risk: Anthropic treats Opus 5.5 as having CB-1 capability (relating to non-novel weapons) but not CB-2 (novel weapons). Requests blocked by the biology classifiers fall back to Opus 5.
• Cyber: Anthropic says it sees "no indication that it can develop novel offensive capabilities," but it is running the cyber classifiers with a temporarily wider safety margin, so some false positives on legitimate security work are likely. In Anthropic's apps, requests blocked by the cyber classifiers fall back to Opus 4.8 automatically. On the API, developers must opt in to fallbacks. Anthropic says Opus 5.5 will come to its Cyber Verification Program, which offers reduced cyber blocking for verified security work, in the near future.
• Fallbacks in practice: in Anthropic's Terminal-Bench-Science run, safeguards flagged 3.9% of requests (affecting 5% of trials), and a fallback model answered them (system card §8.6). If you enable fallbacks, log which model served each request.
• Alignment: on Anthropic's automated behavioral audit, Opus 5.5 showed less misaligned behavior than any other recent Claude model on nearly all measures, and took the fewest overeager or destructive actions of the models tested.
• Regressions Anthropic disclosed: the model is more likely to follow malicious instructions planted in text a user pastes into their own prompt, and it more often accepts unverifiable claims of authorization. In two new evaluations run without safeguards, it tried to escape or tamper with a sandbox in 1.5% of runs. When given apparent credentials to a public package registry in a simulated exercise, it took potentially harmful actions in roughly half of cases.
• Prompt injection: it performed similarly to or better than Opus 5 on every prompt-injection evaluation reported.

In practice: if your agent ingests untrusted pasted content such as issue text, logs or web pages, or holds real credentials, sandbox it and scope its permissions tightly.

Should you switch to Claude Opus 5.5?

• If you are on Opus 5: yes, after a test pass. Opus 5.5 is cheaper per token, scores higher on every row of Anthropic's capability summary, and its fast mode is cheaper and GA. Handle the breaking changes and re-run your effort sweep rather than carrying high or max over.
• If you are on Fable 5.1 for coding agents: test Opus 5.5 at high. It scores higher on CursorBench, SWE-bench Pro and Terminal-Bench, and on AA it beats Fable 5.1 (max) at well under a quarter of the cost per task. Keep Fable 5.1 where it still leads, such as office-document QA. Our Fable 5.1 guide covers where it wins.
• If you are cost-sensitive or need low latency: look at Sonnet 5.5 and Haiku 5.5 first. Sonnet 5.5 costs half Opus 5.5's per-token price, with $0.10 cache reads. Haiku 5.5 starts at $0.10/$0.50. Neither offers fast mode, but Anthropic rates both as faster than Opus 5.5 at standard speed.
• If you run on GCP and want fast mode or day-zero features: evaluate the Anthropic-managed Google Cloud offering instead of the Google-managed route.
• If your pipeline relies on forced tool calls or temperature: budget migration time. You will need strict tool use or structured outputs in place of tool_choice: any, and on API 2026-10-21 you must remove sampling parameters.

FAQ

What is the Claude Opus 5.5 model ID?

It is claude-opus-5-5 on the Claude API, Claude Platform on AWS, both Google Cloud offerings and Microsoft Foundry. On Amazon Bedrock it is anthropic.claude-opus-5-5. There is no date-suffixed snapshot ID.

How much does Claude Opus 5.5 cost?

As of October 2026, it costs $4 per million input tokens and $20 per million output tokens. Cache reads are $0.20 per million (0.05x input), batch is $2/$10, and fast mode is $8/$40. There is no surcharge for long context up to 1M tokens. US-only inference adds 10%.

Is Claude Opus 5.5 fast mode generally available?

Yes. Since 15 October 2026, fast mode on Opus 5.5 needs no beta header or access request. Set speed: "fast". It runs on the Claude API, Microsoft Foundry (Azure-hosted deployments) and, from 22 October, Claude Models on Google Cloud (Anthropic managed). Fast mode on Opus 5 and Opus 4.8 remains a research preview at $10/$50.

How much faster is Opus 5.5 fast mode?

Anthropic says it delivers up to 2.5x higher output tokens per second. The gain is in output throughput, not time to first token, so it shows up most on long streamed responses. It costs twice the standard rate.

Is Claude Opus 5.5 cheaper to run than Opus 5?

Per token, yes: $4/$20 against $5/$25, with cache reads at $0.20 instead of $0.50. Anthropic says it is 40% cheaper on typical workloads at default settings. Opus 5.5 also thinks more per turn at a given effort level, especially at xhigh and max, so the saving can vanish at max effort. Artificial Analysis measured $5.98 per index task for Opus 5.5 (max) against $5.86 for Opus 5 (max), because Opus 5.5 emits far more output tokens. The per-token saving shows up at xhigh and below: at high, Opus 5.5 costs $1.82 per task for a score of 54.

Is Claude Opus 5.5 better than Claude Fable 5.1?

On Anthropic's published benchmarks, mostly yes. It leads Fable 5.1 on every row of the system card's capability summary, including SWE-bench Pro, Terminal-Bench 4.0, Humanity's Last Exam and OSWorld 2.0, plus CursorBench, at 40% of the per-token price. Fable 5.1 still leads on OfficeQA and OfficeQA Pro, and Anthropic says the real-world gap is narrower than the scores suggest.

Does Claude Opus 5.5 have a SWE-bench Verified score?

Not from Anthropic. The announcement and system card report SWE-bench Pro (89.9%), SWE-bench Multilingual (93.9%) and SWE-bench Multimodal (61.4%) instead. Treat any SWE-bench Verified number for Opus 5.5 as third-party.

Can you turn off thinking on Claude Opus 5.5?

No. Adaptive thinking is always on, and sending thinking: {"type": "disabled"} or a manual budget_tokens returns a 400 error. Control thinking depth with the effort parameter instead.

Are Claude Sonnet 5.5 and Haiku 5.5 out?

Yes. Sonnet 5.5 (claude-sonnet-5-5) launched on 28 September 2026 at $2/$10. Haiku 5.5 (claude-haiku-5-5) launched on 7 October 2026 with a 1M context window, starting at $0.10/$0.50. See our Sonnet 5.5 guide and Haiku 5.5 guide.

What is the Claude Opus 5.5 context window?

It is 1M tokens, roughly 555k English words on the current tokenizer. Maximum output is 128K tokens on the Messages API, or 300K on the Message Batches API with the output-300k-2026-03-24 beta header.

Sources

• Claude Platform release notes
• Claude docs: Fast mode
• Claude API pricing
• Claude Opus 5.5 model page
• What's new in Claude Opus 5.5
• Migrating to API version 2026-10-21
• Claude Models on Google Cloud (Anthropic managed)
• Claude docs: Claude Models on Google Cloud (Google managed)
• Anthropic: Introducing Claude Opus 5.5
• Claude Opus 5.5 system card (22 September 2026)
• Artificial Analysis: model leaderboard
• Artificial Analysis: Claude Opus 5.5
• Artificial Analysis: Claude Opus 5

For how Opus 5.5 fits into agent tooling more broadly, see our guide to sandboxing AI coding agents.

If your team is building agents or products on Opus 5.5, Codersera can help you hire vetted remote AI developers who already work with these APIs and know how to tune them for cost and latency.