Anthropic's Python SDK v1.0 swaps httpx. Pin your SDKs.
Anthropic shipped Python SDK v1.0: httpx to httpx2, Python 3.10+, and deprecated surface removed. A reminder that vendor SDK upgrades are a scheduled line item.
A major version bump on a vendor SDK is the kind of thing that breaks a Friday deploy for a team that never opted into it. Anthropic released v1.0 of the Claude Python SDK on August 20, moving the HTTP layer off httpx and dropping years of deprecated surface. Nothing here is unreasonable. All of it will bite you if your requirements.txt says anthropic with no version constraint.
What actually happened
Per the Claude Platform release notes, v1.0 moves the SDK's HTTP layer from httpx to httpx2, which Anthropic describes as a maintained, API-compatible fork. Practically: build custom http_client, Timeout, and transport objects from httpx2 rather than httpx. The DefaultHttpxClient helpers are unchanged. If you use tracing or mocking libraries that monkey-patch httpx — and if you have tests, you probably do — call httpx2.alias_httpx() at startup.
The removals are the sharper edge. v1.0 requires Python 3.10 or later and drops long-deprecated surface: the legacy Text Completions API, and the temperature, top_p, and top_k parameters on Messages methods. The tool runner's client-side compaction_control is gone. On the async client, .with_raw_response results now require await response.parse(). And AnthropicBedrock raises an error when no AWS region is configured instead of quietly defaulting to us-east-1.
That last one is the good kind of breaking change. A silent region default is a bill you find out about later, in a region you did not intend to run in. Anthropic publishes a v1 migration guide with before-and-after snippets for every change.
Why SDK versioning matters for your business
Unpinned dependencies are an outage waiting for a trigger. If your container build runs pip install anthropic fresh, the next image you push picks up v1.0 whether you planned for it or not. Pin the version, upgrade on your calendar, in a branch, with tests.
The sampling-parameter removal is not just cosmetic. temperature and top_p were already rejected on the newer Claude models. Now the SDK will not even accept them. Code carrying those arguments forward from a 2024 prototype fails at the call site instead of the API — which is faster feedback, but still a failure.
Python 3.10+ is a real gate for older deployments. If you have a Lambda, an ECS task, or a customer-hosted box still on 3.9, the SDK upgrade is now a runtime upgrade. Budget accordingly, or stay on 0.x and set a date.
Wrap the client. We put every vendor SDK behind a thin internal module. When a major version lands, one file changes and the migration is a diff a reviewer can actually read — instead of forty call sites and a hope.
Key takeaways
- Claude Python SDK v1.0 released August 20, 2026; HTTP layer moves from
httpxtohttpx2 - Build custom
http_client,Timeout, and transports fromhttpx2;DefaultHttpxClienthelpers unchanged - Call
httpx2.alias_httpx()at startup if tracing or mocking libraries patchhttpx - Requires Python 3.10+; removes Text Completions, and
temperature/top_p/top_kon Messages methods - Async
.with_raw_responsenow needsawait response.parse() AnthropicBedrockerrors on missing AWS region instead of defaulting tous-east-1- Pin the version, upgrade deliberately, and keep vendor SDKs behind one internal wrapper
Your AI stack is a dependency graph, and someone has to own it. We build systems where vendor SDKs sit behind your own interface, versions are pinned, and a major bump is a scheduled branch instead of a broken deploy. See how we build and maintain systems, or have us audit what your builds are pulling in.
Sources: Claude Platform release notes, Anthropic Python SDK v1 migration guide.
- #python
- #claude-api
- #dependency-management
- #sdk
- #technical-debt
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Tricentis AgentScore: score the agent before you ship it
Tricentis is moving QA from deterministic tests to probabilistic agent scoring. The idea is right even if you never buy the product — here's how to build it.
Read itSlack Code: bring your own agent, keep the review loop
Salesforce put Claude Code, Devin, Copilot and Vercel's agent inside Slack channels with diffs, previews and a human approval gate. The interesting part isn't the agents.
Read it