Add no-thinking model picker variants

README.md

@@ -305,6 +305,8 @@ Claude Code 2.1.126 or later reads this proxy's `/v1/models` endpoint when `ANTH
 
 The proxy lists models for configured provider keys and referenced local providers. Picker-safe IDs are routed back to the real provider/model automatically, so no `.env` edit or separate launcher script is needed after startup.
 
+Each provider model also has a `(no thinking)` picker variant. Use it when a model does not support Claude Code thinking or fails with adaptive-thinking requests. It routes to the same upstream model while asking Claude Code to send a non-thinking request.
+
 ## Optional Integrations
 
 ### Discord And Telegram Bots

api/gateway_model_ids.py

@@ -0,0 +1,54 @@
+"""Gateway-safe model id encoding for Claude Code model discovery."""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+GATEWAY_MODEL_ID_PREFIX = "anthropic"
+
+# Claude Code currently treats any model id containing ``claude-3-`` as not
+# supporting thinking. This intentionally uses that client-side capability
+# heuristic while keeping the real provider/model ref reversible for routing.
+NO_THINKING_GATEWAY_MODEL_ID_PREFIX = "claude-3-freecc-no-thinking"
+
+
+@dataclass(frozen=True, slots=True)
+class DecodedGatewayModelId:
+    provider_id: str
+    provider_model: str
+    force_thinking_enabled: bool | None = None
+
+
+def gateway_model_id(provider_model_ref: str) -> str:
+    """Return the normal Claude Code-discoverable id for a provider/model ref."""
+    return f"{GATEWAY_MODEL_ID_PREFIX}/{provider_model_ref}"
+
+
+def no_thinking_gateway_model_id(provider_model_ref: str) -> str:
+    """Return a Claude Code-discoverable id that disables client thinking."""
+    return f"{NO_THINKING_GATEWAY_MODEL_ID_PREFIX}/{provider_model_ref}"
+
+
+def decode_gateway_model_id(model_name: str) -> DecodedGatewayModelId | None:
+    """Decode a model id advertised by this gateway, if it is one."""
+    prefix, separator, remainder = model_name.partition("/")
+    if not separator:
+        return None
+
+    force_thinking_enabled: bool | None
+    if prefix == GATEWAY_MODEL_ID_PREFIX:
+        force_thinking_enabled = None
+    elif prefix == NO_THINKING_GATEWAY_MODEL_ID_PREFIX:
+        force_thinking_enabled = False
+    else:
+        return None
+
+    provider_id, provider_separator, provider_model = remainder.partition("/")
+    if not provider_separator or not provider_model:
+        return None
+
+    return DecodedGatewayModelId(
+        provider_id=provider_id,
+        provider_model=provider_model,
+        force_thinking_enabled=force_thinking_enabled,
+    )

api/model_router.py

@@ -9,10 +9,9 @@ from loguru import logger
 from config.provider_ids import SUPPORTED_PROVIDER_IDS
 from config.settings import Settings
 
+from .gateway_model_ids import decode_gateway_model_id
 from .models.anthropic import MessagesRequest, TokenCountRequest
 
-GATEWAY_MODEL_ID_PREFIX = "anthropic"
-
 
 @dataclass(frozen=True, slots=True)
 class ResolvedModel:
@@ -42,16 +41,23 @@ class ModelRouter:
         self._settings = settings
 
     def resolve(self, claude_model_name: str) -> ResolvedModel:
-        direct_provider_id, direct_provider_model = self._direct_provider_model(
-            claude_model_name
-        )
+        (
+            direct_provider_id,
+            direct_provider_model,
+            force_thinking_enabled,
+        ) = self._direct_provider_model(claude_model_name)
         if direct_provider_id is not None and direct_provider_model is not None:
-            thinking_enabled = self._settings.resolve_thinking(direct_provider_model)
+            thinking_enabled = (
+                force_thinking_enabled
+                if force_thinking_enabled is not None
+                else self._settings.resolve_thinking(direct_provider_model)
+            )
             logger.debug(
-                "MODEL DIRECT: '{}' -> provider='{}' model='{}'",
+                "MODEL DIRECT: '{}' -> provider='{}' model='{}' thinking={}",
                 claude_model_name,
                 direct_provider_id,
                 direct_provider_model,
+                thinking_enabled,
             )
             return ResolvedModel(
                 original_model=claude_model_name,
@@ -77,29 +83,27 @@ class ModelRouter:
             thinking_enabled=thinking_enabled,
         )
 
-    def _direct_provider_model(self, model_name: str) -> tuple[str | None, str | None]:
-        provider_id, separator, provider_model = model_name.partition("/")
-        if not separator:
-            return None, None
-        if provider_id == GATEWAY_MODEL_ID_PREFIX:
-            return self._gateway_encoded_provider_model(provider_model)
-        if provider_id not in SUPPORTED_PROVIDER_IDS:
-            return None, None
-        if not provider_model:
-            return None, None
-        return provider_id, provider_model
-
-    def _gateway_encoded_provider_model(
+    def _direct_provider_model(
         self, model_name: str
-    ) -> tuple[str | None, str | None]:
+    ) -> tuple[str | None, str | None, bool | None]:
+        decoded = decode_gateway_model_id(model_name)
+        if decoded is not None:
+            if decoded.provider_id not in SUPPORTED_PROVIDER_IDS:
+                return None, None, None
+            return (
+                decoded.provider_id,
+                decoded.provider_model,
+                decoded.force_thinking_enabled,
+            )
+
         provider_id, separator, provider_model = model_name.partition("/")
         if not separator:
-            return None, None
+            return None, None, None
         if provider_id not in SUPPORTED_PROVIDER_IDS:
-            return None, None
+            return None, None, None
         if not provider_model:
-            return None, None
-        return provider_id, provider_model
+            return None, None, None
+        return provider_id, provider_model, None
 
     def resolve_messages_request(
         self, request: MessagesRequest

api/routes.py

@@ -9,6 +9,7 @@ from providers.registry import ProviderRegistry
 
 from . import dependencies
 from .dependencies import get_settings, require_api_key
+from .gateway_model_ids import gateway_model_id, no_thinking_gateway_model_id
 from .models.anthropic import MessagesRequest, TokenCountRequest
 from .models.responses import ModelResponse, ModelsListResponse
 from .services import ClaudeProxyService
@@ -16,7 +17,6 @@ from .services import ClaudeProxyService
 router = APIRouter()
 
 DISCOVERED_MODEL_CREATED_AT = "1970-01-01T00:00:00Z"
-GATEWAY_MODEL_ID_PREFIX = "anthropic"
 
 
 SUPPORTED_CLAUDE_MODELS = [
@@ -77,10 +77,6 @@ def _probe_response(allow: str) -> Response:
     return Response(status_code=204, headers={"Allow": allow})
 
 
-def _gateway_model_id(provider_model_ref: str) -> str:
-    return f"{GATEWAY_MODEL_ID_PREFIX}/{provider_model_ref}"
-
-
 def _discovered_model_response(model_id: str, *, display_name: str) -> ModelResponse:
     return ModelResponse(
         id=model_id,
@@ -98,6 +94,27 @@ def _append_unique_model(
     models.append(model)
 
 
+def _append_provider_model_variants(
+    models: list[ModelResponse], seen: set[str], provider_model_ref: str
+) -> None:
+    _append_unique_model(
+        models,
+        seen,
+        _discovered_model_response(
+            gateway_model_id(provider_model_ref),
+            display_name=provider_model_ref,
+        ),
+    )
+    _append_unique_model(
+        models,
+        seen,
+        _discovered_model_response(
+            no_thinking_gateway_model_id(provider_model_ref),
+            display_name=f"{provider_model_ref} (no thinking)",
+        ),
+    )
+
+
 def _build_models_list_response(
     settings: Settings, provider_registry: ProviderRegistry | None
 ) -> ModelsListResponse:
@@ -105,23 +122,11 @@ def _build_models_list_response(
     seen: set[str] = set()
 
     for ref in settings.configured_chat_model_refs():
-        _append_unique_model(
-            models,
-            seen,
-            _discovered_model_response(
-                _gateway_model_id(ref.model_ref), display_name=ref.model_ref
-            ),
-        )
+        _append_provider_model_variants(models, seen, ref.model_ref)
 
     if provider_registry is not None:
         for model_ref in provider_registry.cached_prefixed_model_refs():
-            _append_unique_model(
-                models,
-                seen,
-                _discovered_model_response(
-                    _gateway_model_id(model_ref), display_name=model_ref
-                ),
-            )
+            _append_provider_model_variants(models, seen, model_ref)
 
     for model in SUPPORTED_CLAUDE_MODELS:
         _append_unique_model(models, seen, model)

tests/api/test_model_listing.py

@@ -34,18 +34,29 @@ def test_models_list_includes_configured_refs_cached_provider_models_and_aliases
     data = response.json()
     ids = [item["id"] for item in data["data"]]
 
-    assert ids[:3] == [
+    assert ids[:6] == [
         "anthropic/deepseek/deepseek-chat",
+        "claude-3-freecc-no-thinking/deepseek/deepseek-chat",
         "anthropic/open_router/anthropic/claude-opus",
+        "claude-3-freecc-no-thinking/open_router/anthropic/claude-opus",
         "anthropic/open_router/meta/llama-3.3",
+        "claude-3-freecc-no-thinking/open_router/meta/llama-3.3",
     ]
     assert ids.count("anthropic/deepseek/deepseek-chat") == 1
+    assert ids.count("claude-3-freecc-no-thinking/deepseek/deepseek-chat") == 1
     assert ids.count("anthropic/open_router/anthropic/claude-opus") == 1
+    assert (
+        ids.count("claude-3-freecc-no-thinking/open_router/anthropic/claude-opus") == 1
+    )
     display_names = {item["id"]: item["display_name"] for item in data["data"]}
     assert (
         display_names["anthropic/open_router/meta/llama-3.3"]
         == "open_router/meta/llama-3.3"
     )
+    assert (
+        display_names["claude-3-freecc-no-thinking/open_router/meta/llama-3.3"]
+        == "open_router/meta/llama-3.3 (no thinking)"
+    )
     assert "claude-sonnet-4-20250514" in ids
     assert data["first_id"] == ids[0]
     assert data["last_id"] == ids[-1]
@@ -64,8 +75,10 @@ def test_models_list_works_without_provider_registry():
 
     assert response.status_code == 200
     ids = [item["id"] for item in response.json()["data"]]
-    assert ids[:2] == [
+    assert ids[:4] == [
         "anthropic/deepseek/deepseek-chat",
+        "claude-3-freecc-no-thinking/deepseek/deepseek-chat",
         "anthropic/open_router/anthropic/claude-opus",
+        "claude-3-freecc-no-thinking/open_router/anthropic/claude-opus",
     ]
     assert "claude-sonnet-4-20250514" in ids

tests/api/test_model_router.py

@@ -131,6 +131,27 @@ def test_model_router_routes_gateway_encoded_provider_model_directly(settings):
     )
 
 
+def test_model_router_routes_no_thinking_gateway_model_directly(settings):
+    settings.enable_model_thinking = True
+
+    routed = ModelRouter(settings).resolve_messages_request(
+        MessagesRequest(
+            model="claude-3-freecc-no-thinking/nvidia_nim/deepseek-ai/deepseek-v4-pro",
+            max_tokens=100,
+            messages=[Message(role="user", content="hello")],
+        )
+    )
+
+    assert routed.request.model == "deepseek-ai/deepseek-v4-pro"
+    assert (
+        routed.resolved.original_model
+        == "claude-3-freecc-no-thinking/nvidia_nim/deepseek-ai/deepseek-v4-pro"
+    )
+    assert routed.resolved.provider_id == "nvidia_nim"
+    assert routed.resolved.provider_model == "deepseek-ai/deepseek-v4-pro"
+    assert routed.resolved.thinking_enabled is False
+
+
 def test_model_router_direct_prefixed_model_uses_provider_model_for_thinking(settings):
     settings.enable_model_thinking = False
     settings.enable_opus_thinking = True