Anthropic released Claude Opus 5.5 this week, and it's available on Amazon Web Services, Google Cloud, and Microsoft Azure from day one. Google has also published its own model page for it. I spent some time going through the announcement and the Google Cloud docs, and this release stands out to me for one simple reason: it's more capable and cheaper at the same time. Here's what caught my attention and what you should know before you try it.
Opus 5.5 is the first model in the new Claude 5.5 family. Anthropic says it performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5. That's notable because Fable sits in Anthropic's higher, more expensive tier.
The benchmark numbers back this up. On Terminal-Bench 4.0, which tests multi-step tasks in a command line, Opus 5.5 scores 66.4%, compared with 55.8% for Fable 5.1 and 52.3% for Opus 5. Anthropic does add a fair caveat: at this level of capability, benchmark margins have become a less reliable guide to real-world differences.
Input and output tokens now cost $4 and $20 per million, 20% less than Opus 5. Cache reads, which make up most of the cost in agentic and coding work, are $0.20 per million tokens, 60% cheaper than before. It also generates output more than 30% faster than Opus 5.
The savings add up because it also uses fewer tokens per task. Lower price per token plus fewer tokens nets out to a 40% cost drop.
One note: these are Anthropic's list prices. Google Cloud bills Claude usage through its own generative AI pricing page, so check the rates for your region before estimating costs.
Anthropic says Opus 5.5 is especially good at long, sprawling jobs like codebase-wide migrations and audits. A few examples from the announcement:
This is a quieter change, but I think developers will appreciate it. Anthropic says Opus 5.5 puts the most important information first, is less likely to use jargon, and follows the writing rules you give it. If you've ever tried to review a long, rambling agent summary at the end of the day, you'll know why this matters.
If your team already runs on Google Cloud, this is the easiest way in. You keep your existing IAM, billing, and monitoring, and you get data residency options: multi-region endpoints that route dynamically within the US or EU, or regional endpoints that guarantee routing through a specific region.
It's also useful for planning that Google lists the model's retirement date as no sooner than September 22, 2027, so you can build on it knowing it won't disappear anytime soon.
Getting started looks straightforward. First, install the SDK with Google Cloud support:
pip install -U "anthropic[vertex]"
Then enable the model in Model Garden, authenticate with Application Default Credentials (gcloud auth application-default login), and make a call:
import os
from anthropic import AnthropicVertex
client = AnthropicVertex(
project_id=os.environ["GCP_PROJECT_ID"],
region="us", # multi-region: "us" or "eu"
)
message = client.messages.create(
model="claude-opus-5-5",
max_tokens=1024,
messages=[{"role": "user", "content": "Hey Claude!"}],
)
print(message.content[0].text)
If you call the REST API directly instead of using the SDK, there are two differences from Anthropic's own API: the model is specified in the endpoint URL rather than the request body, and anthropic_version goes in the body with the value vertex-2023-10-16. Anthropic's Claude on Google Cloud guide walks through the full setup.
Swapping the model ID is the easy part. The bigger work is usually fitting a new model into the systems around it: prompts, evals, cost monitoring, and fallbacks. That's the area our team focuses on in LLM development and integration, and it's where I'd spend most of the testing time before any switch.
Anthropic says Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks, with many of the same improvements to performance, efficiency, and safety. If those smaller models get similar efficiency gains, it'll be a very good few months for anyone building with AI.
I'm planning to try Opus 5.5 on our own workloads soon, and I'll share what I find in a follow-up post. Have you tried it yet? Let me know in the comments what you're building with it.