fix(models): fall back to 4096 max_tokens for default Anthropic config - #4373

Related Issue

Closes #4370

Description

Summary

AnthropicModel sends max_tokens=None whenever the user config omits max_tokens, so every call made with the default AnthropicConfig fails with a 400 invalid_request_error from the Anthropic API.

Why (root cause)

PR #3874 changed the max_tokens fallback in AnthropicModel._run/_arun from 4096 to None (camel/models/anthropic_model.py:1156 and camel/models/anthropic_model.py:1301). max_tokens is a required parameter of the Anthropic Messages API: in the pinned anthropic 0.89.0 SDK, messages.create() declares max_tokens: int and enforces it with @required_args, and an explicit None is serialized into the request body as "max_tokens": null rather than being omitted. With the default AnthropicConfig, whose as_dict() drops None values (camel/configs/base_config.py:86), model_config_dict is {}, so the None fallback always applies and every default-config request is rejected.

The goal of #3874 — not capping responses when the user configures their own max_tokens — is preserved: an explicitly configured max_tokens is still passed through unchanged. Only the unset case falls back to an integer again (4096, the value used before #3874 and valid for every Claude model).

Changes

• camel/models/anthropic_model.py: restore the integer fallback for max_tokens in _run and _arun.
• test/models/test_anthropic_model.py: add test_run_default_config_sends_int_max_tokens and test_arun_default_config_sends_int_max_tokens, which construct AnthropicModel with the default config (empty model_config_dict) and mock clients, and assert the outgoing request carries an int max_tokens.

Testing (Python 3.12.14, anthropic 0.89.0, pytest 7.4.4)

New regression tests on the unpatched tree (fail as expected):

$ .venv/bin/python -m pytest test/models/test_anthropic_model.py -k "default_config_sends_int_max_tokens"
>       assert isinstance(request_kwargs["max_tokens"], int)
E       assert False
E        +  where False = isinstance(None, int)
FAILED test/models/test_anthropic_model.py::test_run_default_config_sends_int_max_tokens
FAILED test/models/test_anthropic_model.py::test_arun_default_config_sends_int_max_tokens
======================= 2 failed, 54 deselected in 6.02s =======================

With the fix:

$ .venv/bin/python -m pytest test/models/test_anthropic_model.py
============================== 56 passed in 0.93s ==============================

$ .venv/bin/python -m pytest --fast-test-mode test/models/test_anthropic_model.py
46 passed, 10 skipped in 1.63s

$ .venv/bin/python -m pytest test/models/ --ignore=test/models/test_gemini_model.py --ignore=test/models/reward/test_nemotron_model.py
175 failed, 384 passed, 10 skipped

(The two ignored files fail collection in this environment for missing optional heavy dependencies google-genai/torch; the 175 failures are pre-existing "Missing API key" errors unrelated to this change — the same run on the unpatched tree gives 177 failed / 382 passed, the difference being exactly the two new tests.)

$ .venv/bin/python -m ruff check --no-fix camel/models/anthropic_model.py test/models/test_anthropic_model.py
All checks passed!

$ .venv/bin/python -m ruff format --check camel/models/anthropic_model.py test/models/test_anthropic_model.py
2 files already formatted

$ .venv/bin/python -m mypy --namespace-packages camel/models/anthropic_model.py test/models/test_anthropic_model.py
Success: no issues found in 2 source files

What is the purpose of this pull request?

• Bug fix
• New Feature
• Documentation update
• Other

Checklist

• I have read and agree to the AI-Generated Code Policy (required)
• I have linked this PR to an issue (required) — closes [BUG] AnthropicModel sends max_tokens=None whenever the user config omits it, violating the SDK Required[int] contract, so every default-config call fails #4370
• I have checked if any dependencies need to be added or updated in pyproject.toml and run uv lock (no dependency changes needed)
• I have updated the tests accordingly
• I have updated the documentation if needed (no documentation changes needed)
• I have added examples if this is a new feature (not a new feature)