fix(anthropic): strip vertex @version suffix in model-map candidates so opus 4.7/4.8 capabilities resolve (#30101)

vertex appends @<version> (@default, @20251101) to the routed model id;
the bare name is what carries supports_adaptive_thinking + supports_output_config
in the model map. _model_map_lookup_candidates only stripped bedrock/vertex
prefixes, so vertex_ai/claude-opus-4-8@default never reached claude-opus-4-8.
downstream callers fell through to the name-substring fallback chain which
had no 4.8 case at all.

- _model_map_lookup_candidates: add the suffix-stripped candidate (and the
  prefix-stripped variant of it).
- _is_claude_4_8_model: new helper mirroring 4.7 substring style.
- _is_adaptive_thinking_model: include 4.8 in the name-fallback OR chain
  for defence in depth.

Fixes #30101
This commit is contained in:
Chenglun Hu 2026-06-10 16:30:44 +08:00
parent e15b37a18e
commit 21aec374d9
2 changed files with 101 additions and 3 deletions

View file

@ -3,6 +3,7 @@ This file contains common utils for anthropic calls.
"""
import copy
import re
from typing import Any, Dict, List, Optional, Union
import httpx
@ -271,6 +272,20 @@ class AnthropicModelInfo(BaseLLMModelInfo):
)
)
@staticmethod
def _is_claude_4_8_model(model: str) -> bool:
"""Check if the model is a Claude 4.8 model (Opus 4.8)."""
model_lower = model.lower()
return any(
v in model_lower
for v in (
"opus-4-8",
"opus_4_8",
"opus-4.8",
"opus_4.8",
)
)
@staticmethod
def _supports_sampling_params(model: str) -> bool:
"""Claude 4.7+ (Opus 4.7/4.8, Fable 5) removed sampling params: the API
@ -335,6 +350,15 @@ class AnthropicModelInfo(BaseLLMModelInfo):
"""Model-map keys to try for ``model``, stripping bedrock/vertex
prefixes so a provider-routed Claude still resolves to its entry."""
candidates = [model]
# Vertex appends ``@<version>`` (e.g. ``@default``, ``@20251101``) to
# the model id; the bare name is what carries the capability flags in
# the model map. Without this strip, ``claude-opus-4-8@default``
# misses ``claude-opus-4-8`` and downstream
# ``supports_adaptive_thinking`` / ``supports_output_config`` checks
# silently fall back to the partial name-substring path (#30101).
suffix_stripped = re.sub(r"@[^@/]+$", "", model)
if suffix_stripped != model:
candidates.append(suffix_stripped)
for prefix in (
"bedrock/converse/",
"bedrock/invoke/",
@ -343,6 +367,8 @@ class AnthropicModelInfo(BaseLLMModelInfo):
):
if model.startswith(prefix):
candidates.append(model[len(prefix) :])
if suffix_stripped != model and suffix_stripped.startswith(prefix):
candidates.append(suffix_stripped[len(prefix) :])
try:
from litellm.llms.bedrock.common_utils import BedrockModelInfo
@ -399,9 +425,11 @@ class AnthropicModelInfo(BaseLLMModelInfo):
model, "supports_adaptive_thinking"
):
return True
return AnthropicModelInfo._is_claude_4_6_model(
model
) or AnthropicModelInfo._is_claude_4_7_model(model)
return (
AnthropicModelInfo._is_claude_4_6_model(model)
or AnthropicModelInfo._is_claude_4_7_model(model)
or AnthropicModelInfo._is_claude_4_8_model(model)
)
def is_effort_used(
self, optional_params: Optional[dict], model: Optional[str] = None

View file

@ -1450,3 +1450,73 @@ class TestClaudeOpus48AdaptiveThinking:
from litellm.llms.anthropic.common_utils import AnthropicModelInfo
assert AnthropicModelInfo._is_adaptive_thinking_model(model) is False
class TestVertexVersionSuffixModelResolution:
"""Regression for #30101.
Vertex appends ``@<version>`` (``@default``, ``@20251101``, ...) to the
routed model id. ``_model_map_lookup_candidates`` stripped bedrock/vertex
prefixes but not the trailing suffix, so a request id like
``vertex_ai/claude-opus-4-8@default`` never resolved to the bare
``claude-opus-4-8`` map entry that carries the ``supports_adaptive_thinking``
and ``supports_output_config`` flags. Downstream adaptive-thinking +
effort detection fell through to the partial name-substring path.
"""
@pytest.mark.parametrize(
"model",
[
"claude-opus-4-7@default",
"claude-opus-4-8@default",
"claude-opus-4-8@20251101",
"vertex_ai/claude-opus-4-7@default",
"vertex_ai/claude-opus-4-8@default",
],
)
def test_adaptive_thinking_detected_with_vertex_version_suffix(
self, local_model_cost_map, model
):
from litellm.llms.anthropic.common_utils import AnthropicModelInfo
assert AnthropicModelInfo._is_adaptive_thinking_model(model) is True
@pytest.mark.parametrize(
"model",
[
"claude-opus-4-7@default",
"claude-opus-4-8@default",
"vertex_ai/claude-opus-4-8@default",
],
)
def test_capability_resolves_through_version_suffix(
self, local_model_cost_map, model
):
from litellm.llms.anthropic.common_utils import AnthropicModelInfo
assert (
AnthropicModelInfo._supports_model_capability(
model, "supports_adaptive_thinking"
)
is True
)
assert (
AnthropicModelInfo._supports_model_capability(
model, "supports_output_config"
)
is True
)
def test_is_claude_4_8_model_recognizes_suffix_and_provider_prefix(self):
from litellm.llms.anthropic.common_utils import AnthropicModelInfo
assert AnthropicModelInfo._is_claude_4_8_model("claude-opus-4-8") is True
assert (
AnthropicModelInfo._is_claude_4_8_model("claude-opus-4-8@default") is True
)
assert (
AnthropicModelInfo._is_claude_4_8_model("vertex_ai/claude-opus-4-8@default")
is True
)
assert AnthropicModelInfo._is_claude_4_8_model("claude-opus-4-7") is False
assert AnthropicModelInfo._is_claude_4_8_model("claude-opus-4-5") is False