Skip to content

Use Claude Code, Codex, VSCode, Pi, and OpenCode (and 6 other harnesses) for free (1.3B+ free tokens) from your terminal, app, IDE, or phone, and now from the browser with native browser sessions (multi-harness + multi-model) like OpenClaw (voice supported + ToS friendly)

Python

Add QwenCloud Coding Plan as a separate provider (#1425)

Ali Khokhar92801b3a36c6
diff --git a/.env.example b/.env.example
index d07ee907..65fb034f 100644
--- a/.env.example
+++ b/.env.example
@@ -112,6 +112,10 @@ VERTEX_LOCATION="global"
# QWENCLOUD_API_KEY=<your-token>
+# QwenCloud Coding Plan (separate sk-sp- key; personal interactive coding-agent use)
+# QWENCLOUD_CODING_API_KEY=<your-token>
+
+
# Together AI (OpenAI-compatible Chat Completions at api.together.ai/v1)
# TOGETHER_API_KEY=<your-token>
@@ -178,7 +182,7 @@ OLLAMA_BASE_URL="http://localhost:11434"
# All Claude model requests are mapped to these models, plain model is fallback
# Format: provider_type/model/name
-# Valid providers: "nvidia_nim" | "openai" | "azure_openai" | "open_router" | "gemini" | "vertex" | "deepseek" | "mistral" | "mistral_codestral" | "opencode_zen" | "opencode_go" | "vercel" | "bedrock" | "huggingface" | "cohere" | "github_models" | "wafer" | "kimi" | "kimi_code" | "minimax" | "cerebras" | "groq" | "xai" | "qwencloud" | "together" | "deepinfra" | "siliconflow" | "nebius" | "chutes" | "featherless" | "agnes" | "zenmux" | "wandb" | "sambanova" | "kilo" | "fireworks" | "novita" | "cloudflare" | "zai" | "ollama_cloud" | "lmstudio" | "llamacpp" | "ollama" | "tokenrouter" | "nararoute"
+# Valid providers: "nvidia_nim" | "openai" | "azure_openai" | "open_router" | "gemini" | "vertex" | "deepseek" | "mistral" | "mistral_codestral" | "opencode_zen" | "opencode_go" | "vercel" | "bedrock" | "huggingface" | "cohere" | "github_models" | "wafer" | "kimi" | "kimi_code" | "minimax" | "cerebras" | "groq" | "xai" | "qwencloud" | "qwencloud_coding" | "together" | "deepinfra" | "siliconflow" | "nebius" | "chutes" | "featherless" | "agnes" | "zenmux" | "wandb" | "sambanova" | "kilo" | "fireworks" | "novita" | "cloudflare" | "zai" | "ollama_cloud" | "lmstudio" | "llamacpp" | "ollama" | "tokenrouter" | "nararoute"
# MODEL_FABLE=<provider/model-id>
# MODEL_OPUS=<provider/model-id>
# MODEL_SONNET=<provider/model-id>
@@ -221,6 +225,7 @@ MODEL="nvidia_nim/nvidia/nemotron-3-super-120b-a12b"
# FCC_SMOKE_MODEL_VERTEX=<provider/model-id>
# FCC_SMOKE_MODEL_GROQ=<provider/model-id>
# FCC_SMOKE_MODEL_QWENCLOUD=<provider/model-id>
+# FCC_SMOKE_MODEL_QWENCLOUD_CODING=<provider/model-id>
# FCC_SMOKE_MODEL_TOGETHER=<provider/model-id>
# FCC_SMOKE_MODEL_DEEPINFRA=<provider/model-id>
# FCC_SMOKE_MODEL_SILICONFLOW=<provider/model-id>
@@ -255,6 +260,7 @@ REASONING_HAIKU=inherit
# OPENAI_PROXY=<http-or-socks-url>
# XAI_PROXY=<http-or-socks-url>
# QWENCLOUD_PROXY=<http-or-socks-url>
+# QWENCLOUD_CODING_PROXY=<http-or-socks-url>
# TOGETHER_PROXY=<http-or-socks-url>
# DEEPINFRA_PROXY=<http-or-socks-url>
# SILICONFLOW_PROXY=<http-or-socks-url>
diff --git a/ARCHITECTURE.md b/ARCHITECTURE.md
index c5fa87b2..78aff29a 100644
--- a/ARCHITECTURE.md
+++ b/ARCHITECTURE.md
@@ -776,6 +776,11 @@ declarative profiles for their thinking, token, and `extra_body` policy. Kimi
Code remains distinct from Kimi API because its subscription key and base URL
are a separate customer contract; its profile maps provider-neutral reasoning
to Kimi's named efforts and identifies FCC through the upstream user agent.
+QwenCloud Coding Plan likewise remains distinct from QwenCloud Token Plan
+because its subscription key, quota, endpoint, and personal interactive-use
+contract are separate. It uses the ordinary OpenAI Chat transport, preserves
+reasoning history through `reasoning_content`, and does not impose one reasoning
+control or output-token default across its heterogeneous model catalog.
Z.ai is treated as the GLM Coding Plan provider and uses Z.ai's Coding Plan
OpenAI base.
Mistral La Plateforme keeps its native `reasoning_effort` and thinking-chunk
diff --git a/README.md b/README.md
index 56f1badf..780d1c32 100644
--- a/README.md
+++ b/README.md
@@ -162,6 +162,7 @@ fcc-codex exec "hello"
| [OpenAI / ChatGPT](https://learn.chatgpt.com/docs/auth) | Connect ChatGPT in the Admin UI | `openai/<model-id>` |
| [xAI (Grok)](https://console.x.ai/team/default/api-keys) | `XAI_API_KEY` | `xai/grok-4.5` |
| [QwenCloud Token Plan](https://home.qwencloud.com/api-keys) | `QWENCLOUD_API_KEY` | `qwencloud/qwen3.7-plus` |
+| [QwenCloud Coding Plan](https://home.qwencloud.com/api-keys) | `QWENCLOUD_CODING_API_KEY` | `qwencloud_coding/qwen3.7-plus` |
| [Together AI](https://api.together.ai/settings/api-keys) | `TOGETHER_API_KEY` | `together/zai-org/GLM-5.2` |
| [DeepInfra](https://deepinfra.com/dash/api_keys) | `DEEPINFRA_API_KEY` | `deepinfra/deepseek-ai/DeepSeek-V4-Flash` |
| [SiliconFlow](https://cloud.siliconflow.com/account/ak) | `SILICONFLOW_API_KEY` | `siliconflow/Qwen/Qwen3-32B` |
@@ -219,6 +220,10 @@ fcc-codex exec "hello"
- Kimi Code subscription keys use `kimi_code/`; Kimi API credit keys use
`kimi/`. Kimi Code plans are for personal interactive coding-agent use under
[Kimi's community guidelines](https://www.kimi.com/code/docs/en/kimi-code/community-guidelines.html).
+- QwenCloud Coding Plan keys use `qwencloud_coding/`; QwenCloud Token Plan keys
+ use `qwencloud/`. The keys and endpoints are not interchangeable. Coding Plan
+ is for local, personal, interactive coding-agent use under the
+ [Coding Plan terms](https://www.alibabacloud.com/help/en/model-studio/coding-plan).
- OpenCode Zen and OpenCode Go share `OPENCODE_API_KEY` but use the explicit
`opencode_zen/` and `opencode_go/` model prefixes.
- For Amazon Bedrock, set `BEDROCK_BASE_URL` to the URL for the same region as
diff --git a/pyproject.toml b/pyproject.toml
index 6500fb84..c8f47271 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
[project]
name = "free-claude-code"
-version = "5.1.2"
+version = "5.2.0"
description = "Local proxy connecting coding agents to OpenAI-compatible AI providers"
readme = "README.md"
requires-python = ">=3.14.0"
diff --git a/smoke/lib/config.py b/smoke/lib/config.py
index 9b24cba9..94a6eb1b 100644
--- a/smoke/lib/config.py
+++ b/smoke/lib/config.py
@@ -73,6 +73,7 @@ PROVIDER_SMOKE_DEFAULT_MODELS: dict[str, str] = {
"groq": "groq/llama-3.3-70b-versatile",
"xai": "xai/grok-4.5",
"qwencloud": "qwencloud/qwen3.7-plus",
+ "qwencloud_coding": "qwencloud_coding/qwen3.7-plus",
"together": "together/zai-org/GLM-5.2",
"deepinfra": "deepinfra/deepseek-ai/DeepSeek-V4-Flash",
"siliconflow": "siliconflow/Qwen/Qwen3-32B",
diff --git a/src/free_claude_code/config/admin/provider_manifest.py b/src/free_claude_code/config/admin/provider_manifest.py
index c16b208b..d9151001 100644
--- a/src/free_claude_code/config/admin/provider_manifest.py
+++ b/src/free_claude_code/config/admin/provider_manifest.py
@@ -152,6 +152,14 @@ _PROVIDER_FIELD_OVERRIDES: dict[str, dict[str, Any]] = {
"mixed."
),
},
+ "QWENCLOUD_CODING_API_KEY": {
+ "label": "QwenCloud Coding Plan API Key",
+ "description": (
+ "Dedicated QwenCloud Coding Plan key (sk-sp-...) for personal, "
+ "interactive coding-agent use. Token Plan, Coding Plan, and "
+ "pay-as-you-go keys use separate endpoints and cannot be mixed."
+ ),
+ },
"TOGETHER_API_KEY": {
"label": "Together AI API Key",
"description": (
diff --git a/src/free_claude_code/config/provider_catalog.py b/src/free_claude_code/config/provider_catalog.py
index 1e3aa1a5..844ee8eb 100644
--- a/src/free_claude_code/config/provider_catalog.py
+++ b/src/free_claude_code/config/provider_catalog.py
@@ -56,6 +56,8 @@ XAI_DEFAULT_BASE = "https://api.x.ai/v1"
QWENCLOUD_DEFAULT_BASE = (
"https://token-plan.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1"
)
+# QwenCloud Coding Plan OpenAI-compatible Chat Completions API.
+QWENCLOUD_CODING_DEFAULT_BASE = "https://coding-intl.dashscope.aliyuncs.com/v1"
# Together AI OpenAI-compatible Chat Completions API.
TOGETHER_DEFAULT_BASE = "https://api.together.ai/v1"
# DeepInfra OpenAI-compatible Chat Completions API.
@@ -168,6 +170,15 @@ PROVIDER_CATALOG: dict[str, ProviderDescriptor] = {
default_base_url=QWENCLOUD_DEFAULT_BASE,
proxy_attr="qwencloud_proxy",
),
+ "qwencloud_coding": ProviderDescriptor(
+ provider_id="qwencloud_coding",
+ display_name="QwenCloud Coding Plan",
+ credential_env="QWENCLOUD_CODING_API_KEY",
+ credential_url="https://home.qwencloud.com/api-keys",
+ credential_attr="qwencloud_coding_api_key",
+ default_base_url=QWENCLOUD_CODING_DEFAULT_BASE,
+ proxy_attr="qwencloud_coding_proxy",
+ ),
"together": ProviderDescriptor(
provider_id="together",
display_name="Together AI",
diff --git a/src/free_claude_code/config/settings.py b/src/free_claude_code/config/settings.py
index 97840002..8b32e832 100644
--- a/src/free_claude_code/config/settings.py
+++ b/src/free_claude_code/config/settings.py
@@ -210,6 +210,11 @@ class Settings(BaseModel):
default=None, validation_alias="QWENCLOUD_API_KEY"
)
+ # ==================== QwenCloud Coding Plan (OpenAI-compatible) ====================
+ qwencloud_coding_api_key: OptionalNonEmptyString = Field(
+ default=None, validation_alias="QWENCLOUD_CODING_API_KEY"
+ )
+
# ==================== Together AI (OpenAI-compatible) ====================
together_api_key: OptionalNonEmptyString = Field(
default=None, validation_alias="TOGETHER_API_KEY"
@@ -334,6 +339,9 @@ class Settings(BaseModel):
qwencloud_proxy: OptionalNonEmptyString = Field(
default=None, validation_alias="QWENCLOUD_PROXY"
)
+ qwencloud_coding_proxy: OptionalNonEmptyString = Field(
+ default=None, validation_alias="QWENCLOUD_CODING_PROXY"
+ )
together_proxy: OptionalNonEmptyString = Field(
default=None, validation_alias="TOGETHER_PROXY"
)
diff --git a/src/free_claude_code/providers/openai_chat/output_cap.py b/src/free_claude_code/providers/openai_chat/output_cap.py
index cfdd4b5d..bd9cf5e8 100644
--- a/src/free_claude_code/providers/openai_chat/output_cap.py
+++ b/src/free_claude_code/providers/openai_chat/output_cap.py
@@ -11,7 +11,6 @@ retry once and succeed. The provider also remembers the learned cap per model
so later requests clamp proactively instead of paying the 400 every time.
"""
-import json
import re
from typing import Any
@@ -19,20 +18,41 @@ import openai
# Body keys that carry the output-token budget across OpenAI-compatible policies.
_OUTPUT_TOKEN_FIELDS = ("max_completion_tokens", "max_tokens")
+_STRUCTURED_TEXT_FIELDS = ("message", "detail", "error")
+_STRUCTURED_CONTAINER_FIELDS = ("error", "errors", "detail", "details")
-# Only treat a 400 as an output-cap rejection when it names one of these fields.
-_OUTPUT_TOKEN_KEYWORDS = ("max_completion_tokens", "max_tokens")
-
-# Comparator phrases that precede the allowed maximum in provider error text.
_CAP_VALUE_PATTERN = r"[`'\"]?(\d+)[`'\"]?"
-_CAP_PATTERNS: tuple[re.Pattern[str], ...] = (
- re.compile(rf"less than or equal to\s+{_CAP_VALUE_PATTERN}"),
- re.compile(rf"smaller than or equal to\s+{_CAP_VALUE_PATTERN}"),
- re.compile(rf"<=\s*{_CAP_VALUE_PATTERN}"),
- re.compile(rf"at most\s+{_CAP_VALUE_PATTERN}"),
- re.compile(rf"must not exceed\s+{_CAP_VALUE_PATTERN}"),
- re.compile(rf"maximum(?:\s+value)?(?:\s+for\s+\S+)?\s+is\s+{_CAP_VALUE_PATTERN}"),
- re.compile(rf"maximum(?:\s+allowed)?(?:\s+value)?\s+of\s+{_CAP_VALUE_PATTERN}"),
+_OUTPUT_TOKEN_FIELD_PATTERN = (
+ r"(?<!\w)[`'\"]?(?:max_completion_tokens|max_tokens)[`'\"]?(?!\w)"
+)
+
+# An accepted grammar must bind the output-token field and its numeric limit.
+# A field mentioned elsewhere in the response must never authorize an unrelated
+# comparator.
+_FIELD_BOUND_CAP_PATTERNS: tuple[re.Pattern[str], ...] = (
+ re.compile(
+ rf"range\s+of\s+{_OUTPUT_TOKEN_FIELD_PATTERN}\s+should\s+be\s+"
+ rf"\[\s*1\s*,\s*{_CAP_VALUE_PATTERN}\s*\]"
+ ),
+ re.compile(
+ rf"{_OUTPUT_TOKEN_FIELD_PATTERN}\s*(?::\s*)?"
+ rf"(?:"
+ rf"(?:(?:must|should)\s+be\s+)?"
+ rf"(?:less|smaller)\s+than\s+or\s+equal\s+to"
+ rf"|(?:(?:must|should)\s+be\s*)?<="
+ rf"|(?:(?:must|should)\s+be\s+)?at\s+most"
+ rf"|(?:must|should)\s+not\s+exceed"
+ rf"|maximum(?:\s+allowed)?(?:\s+value)?\s+(?:is|of)"
+ rf")\s*{_CAP_VALUE_PATTERN}"
+ ),
+ re.compile(
+ rf"maximum(?:\s+allowed)?(?:\s+value)?\s+for\s+"
+ rf"{_OUTPUT_TOKEN_FIELD_PATTERN}\s+is\s+{_CAP_VALUE_PATTERN}"
+ ),
+ re.compile(
+ rf"maximum(?:\s+allowed)?(?:\s+value)?\s+of\s+"
+ rf"{_CAP_VALUE_PATTERN}\s+for\s+{_OUTPUT_TOKEN_FIELD_PATTERN}"
+ ),
)
@@ -42,12 +62,75 @@ def _is_bad_request(error: Exception) -> bool:
)
-def _error_text(error: Exception) -> str:
- text = str(error)
+def _normalized_output_field(value: object) -> str | None:
+ if not isinstance(value, str):
+ return None
+ field = value.strip("`'\" ").lower()
+ return field if field in _OUTPUT_TOKEN_FIELDS else None
+
+
+def _structured_error_texts(body: dict[str, Any] | list[Any]) -> tuple[str, ...]:
+ """Extract messages without discarding their structured parameter scope."""
+ texts: list[str] = []
+ pending: list[dict[str, Any] | list[Any]] = [body]
+ while pending:
+ value = pending.pop()
+ if isinstance(value, list):
+ for item in value:
+ if isinstance(item, str):
+ texts.append(item)
+ elif isinstance(item, (dict, list)):
+ pending.append(item)
+ continue
+
+ param = value.get("param")
+ output_field = _normalized_output_field(param)
+ messages = tuple(
+ message
+ for field in _STRUCTURED_TEXT_FIELDS
+ if isinstance(message := value.get(field), str)
+ )
+ if output_field is not None:
+ # Structured parameter metadata is authoritative. Prefixing the
+ # paired field supports comparator-only messages such as "<= 8192".
+ texts.extend(f"{output_field} {message}" for message in messages)
+ if param is not None:
+ # Do not let text nested below an explicitly scoped validation error
+ # escape that scope and become a new unscoped candidate.
+ continue
+
+ texts.extend(messages)
+ for field in _STRUCTURED_CONTAINER_FIELDS:
+ nested = value.get(field)
+ if isinstance(nested, (dict, list)):
+ pending.append(nested)
+ return tuple(texts)
+
+
+def _error_texts(error: Exception) -> tuple[str, ...]:
+ """Keep unstructured text separate from structured validation messages."""
body = getattr(error, "body", None)
- if body is not None:
- text = f"{text} {json.dumps(body, default=str)}"
- return text.lower()
+ if body is None:
+ texts = (str(error),)
+ elif isinstance(body, str):
+ texts = (body,)
+ elif isinstance(body, (dict, list)):
+ # The OpenAI SDK embeds a representation of this body in ``str(error)``.
+ # Parsing both would flatten the parameter scopes restored above.
+ texts = _structured_error_texts(body)
+ else:
+ texts = ()
+ return tuple(text.lower() for text in texts)
+
+
+def _parse_cap(text: str) -> int | None:
+ for pattern in _FIELD_BOUND_CAP_PATTERNS:
+ match = pattern.search(text)
+ if match:
+ cap = int(match.group(1))
+ if cap > 0:
+ return cap
+ return None
def parse_output_token_cap(error: Exception) -> int | None:
@@ -55,16 +138,10 @@ def parse_output_token_cap(error: Exception) -> int | None:
if not _is_bad_request(error):
return None
- text = _error_text(error)
- if not any(keyword in text for keyword in _OUTPUT_TOKEN_KEYWORDS):
- return None
-
- for pattern in _CAP_PATTERNS:
- match = pattern.search(text)
- if match:
- cap = int(match.group(1))
- if cap > 0:
- return cap
+ for text in _error_texts(error):
+ cap = _parse_cap(text)
+ if cap is not None:
+ return cap
return None
diff --git a/src/free_claude_code/providers/openai_chat/profiles.py b/src/free_claude_code/providers/openai_chat/profiles.py
index 4bc5fced..244bef59 100644
--- a/src/free_claude_code/providers/openai_chat/profiles.py
+++ b/src/free_claude_code/providers/openai_chat/profiles.py
@@ -214,6 +214,13 @@ OPENAI_CHAT_PROFILES: dict[str, OpenAIChatProfile] = {
),
NO_REASONING,
),
+ "qwencloud_coding": OpenAIChatProfile(
+ _policy(
+ "QWENCLOUD_CODING",
+ ReasoningReplayMode.REASONING_CONTENT,
+ ),
+ NO_REASONING,
+ ),
"together": OpenAIChatProfile(
_policy(
"TOGETHER",
diff --git a/tests/contracts/test_admin_provider_manifest.py b/tests/contracts/test_admin_provider_manifest.py
index 0c28a24a..4cf220b8 100644
--- a/tests/contracts/test_admin_provider_manifest.py
+++ b/tests/contracts/test_admin_provider_manifest.py
@@ -139,6 +139,16 @@ def test_vertex_project_and_location_are_admin_provider_fields() -> None:
assert location.resolved_default() == "global"
+def test_qwencloud_coding_key_is_a_distinct_admin_provider_field() -> None:
+ entry = FIELD_BY_KEY["QWENCLOUD_CODING_API_KEY"]
+
+ assert entry.label == "QwenCloud Coding Plan API Key"
+ assert entry.settings_attr == "qwencloud_coding_api_key"
+ assert entry.section_id == "providers"
+ assert entry.secret is True
+ assert "separate endpoints" in entry.description
+
+
def test_vertex_admin_status_uses_project_configuration_not_an_api_key() -> None:
from free_claude_code.config.admin.status import provider_config_status
diff --git a/tests/contracts/test_provider_catalog_order.py b/tests/contracts/test_provider_catalog_order.py
index 621f68c9..639f549b 100644
--- a/tests/contracts/test_provider_catalog_order.py
+++ b/tests/contracts/test_provider_catalog_order.py
@@ -12,6 +12,7 @@ _EXPECTED_PROVIDER_ORDER: tuple[str, ...] = (
"openai",
"xai",
"qwencloud",
+ "qwencloud_coding",
"together",
"deepinfra",
"siliconflow",
diff --git a/tests/contracts/test_smoke_config.py b/tests/contracts/test_smoke_config.py
index 3262d119..1c2ebc91 100644
--- a/tests/contracts/test_smoke_config.py
+++ b/tests/contracts/test_smoke_config.py
@@ -55,6 +55,7 @@ def _settings(**overrides):
"groq_api_key": "",
"xai_api_key": "",
"qwencloud_api_key": "",
+ "qwencloud_coding_api_key": "",
"together_api_key": "",
"deepinfra_api_key": "",
"siliconflow_api_key": "",
@@ -194,6 +195,52 @@ def test_qwencloud_provider_smoke_uses_current_coding_model(monkeypatch) -> None
assert models[0].source == "provider_default"
+def test_qwencloud_coding_provider_smoke_uses_recommended_model(monkeypatch) -> None:
+ monkeypatch.delenv("FCC_SMOKE_MODEL_QWENCLOUD_CODING", raising=False)
+ config = _smoke_config(
+ settings=_settings(
+ model="ollama/llama3.1",
+ ollama_base_url="",
+ qwencloud_coding_api_key="qwencloud-coding-key",
+ )
+ )
+
+ models = config.provider_smoke_models()
+
+ assert [model.provider for model in models] == ["qwencloud_coding"]
+ assert models[0].full_model == "qwencloud_coding/qwen3.7-plus"
+ assert models[0].source == "provider_default"
+
+
+def test_qwencloud_coding_smoke_override_is_independent_from_token_plan(
+ monkeypatch,
+) -> None:
+ monkeypatch.delenv("FCC_SMOKE_MODEL_QWENCLOUD", raising=False)
+ monkeypatch.setenv(
+ "FCC_SMOKE_MODEL_QWENCLOUD_CODING",
+ "qwencloud_coding/kimi-k2.5",
+ )
+ config = _smoke_config(
+ settings=_settings(
+ model="ollama/llama3.1",
+ ollama_base_url="",
+ qwencloud_api_key="qwencloud-key",
+ qwencloud_coding_api_key="qwencloud-coding-key",
+ )
+ )
+
+ models = config.provider_smoke_models()
+
+ assert [(model.provider, model.full_model, model.source) for model in models] == [
+ ("qwencloud", "qwencloud/qwen3.7-plus", "provider_default"),
+ (
+ "qwencloud_coding",
+ "qwencloud_coding/kimi-k2.5",
+ "FCC_SMOKE_MODEL_QWENCLOUD_CODING",
+ ),
+ ]
+
+
def test_together_provider_smoke_uses_current_coding_model(monkeypatch) -> None:
monkeypatch.delenv("FCC_SMOKE_MODEL_TOGETHER", raising=False)
config = _smoke_config(
diff --git a/tests/providers/test_openai_chat_output_cap.py b/tests/providers/test_openai_chat_output_cap.py
index 6127d966..f2a06b94 100644
--- a/tests/providers/test_openai_chat_output_cap.py
+++ b/tests/providers/test_openai_chat_output_cap.py
@@ -67,6 +67,30 @@ def test_parse_cap_reads_json_body():
assert parse_output_token_cap(error) == 12000
+ "message,expected",
+ [
+ ("Range of max_tokens should be [1, 32768]", 32768),
+ ("range of max_completion_tokens should be [ 1 , `8192` ]", 8192),
+ ],
+)
+def test_parse_cap_from_inclusive_range(message, expected):
+ assert parse_output_token_cap(_BadRequest(message)) == expected
+
+
+ "message",
+ [
+ "Range of temperature should be [0, 2] for max_tokens requests",
+ "Range of max_tokens should be [0, 4096]",
+ "Range of max_tokens should be [1, unlimited]",
+ "Range of max_tokens is [1, 4096]",
+ ],
+)
+def test_parse_cap_ignores_unrecognized_ranges(message):
+ assert parse_output_token_cap(_BadRequest(message)) is None
+
+
def test_parse_cap_ignores_non_400():
error = _BadRequest("max_tokens must be less than or equal to 40960")
error.status_code = 500
@@ -77,6 +101,92 @@ def test_parse_cap_ignores_unrelated_400():
assert parse_output_token_cap(_BadRequest("temperature must be <= 2")) is None
+ "message",
+ [
+ "temperature must be <= 2; max_tokens is invalid",
+ "max_completion_tokens is invalid; temperature at most 2",
+ "maximum allowed value of 2 for temperature; max_tokens is invalid",
+ ],
+)
+def test_parse_cap_does_not_bind_unrelated_comparator_to_output_field(message):
+ assert parse_output_token_cap(_BadRequest(message)) is None
+
+
+def test_parse_cap_uses_only_the_structured_output_parameter_message():
+ error = _BadRequest(
+ "max_tokens is invalid",
+ body={
+ "errors": [
+ {"param": "temperature", "message": "must not exceed 2"},
+ {"param": "top_p", "message": "<= 1"},
+ ]
+ },
+ )
+
+ assert parse_output_token_cap(error) is None
+
+
+def test_parse_cap_respects_structured_non_output_parameter():
+ body = {
+ "param": "temperature",
+ "message": "max_tokens must not exceed 2",
+ }
+ error = _BadRequest(
+ f"Error code: 400 - {body}",
+ body=body,
+ )
+
+ assert parse_output_token_cap(error) is None
+
+
+def test_parse_cap_does_not_escape_structured_parameter_scope():
+ error = _BadRequest(
+ "invalid request",
+ body={
+ "param": "temperature",
+ "details": {"message": "max_tokens must not exceed 2"},
+ },
+ )
+
+ assert parse_output_token_cap(error) is None
+
+
+def test_parse_cap_ignores_text_outside_structured_error_schema():
+ error = _BadRequest(
+ "invalid request",
+ body={"request": {"message": "max_tokens must not exceed 2"}},
+ )
+
+ assert parse_output_token_cap(error) is None
+
+
+def test_parse_cap_reads_unscoped_structured_message():
+ error = _BadRequest(
+ "invalid request",
+ body={"message": "max_tokens must not exceed 8192"},
+ )
+
+ assert parse_output_token_cap(error) == 8192
+
+
+def test_parse_cap_selects_matching_parameter_from_structured_error_list():
+ error = _BadRequest(
+ "invalid request",
+ body={
+ "errors": [
+ {
+ "param": "temperature",
+ "message": "max_tokens must not exceed 2",
+ },
+ {"param": "max_tokens", "message": "<= 8192"},
+ ]
+ },
+ )
+
+ assert parse_output_token_cap(error) == 8192
+
+
def test_parse_cap_returns_none_without_number():
assert (
parse_output_token_cap(_BadRequest("max_tokens is larger than allowed")) is None
@@ -213,3 +323,36 @@ async def test_unrelated_400_is_not_clamped_and_propagates(groq_provider):
assert create.call_count == 1
assert groq_provider._model_output_caps == {}
+
+
+async def test_mixed_field_400_does_not_retry_or_poison_learned_cap(groq_provider):
+ body = groq_provider._build_request_body(
+ make_messages_request(
+ "llama-3.3-70b-versatile",
+ max_tokens=64000,
+ thinking={"enabled": False},
+ )
+ )
+ error_body = {
+ "param": "temperature",
+ "message": "max_completion_tokens must not exceed 2",
+ }
+ create = AsyncMock(
+ side_effect=_BadRequest(
+ f"Error code: 400 - {error_body}",
+ body=error_body,
+ )
+ )
+
+ with (
+ patch.object(groq_provider._client.chat.completions, "create", create),
+ pytest.raises(Exception, match="temperature"),
+ ):
+ await groq_provider._create_stream(
+ body,
+ groq_provider._admission.new_retry_session(),
+ )
+
+ assert create.call_count == 1
+ assert groq_provider._model_output_caps == {}
diff --git a/tests/providers/test_provider_runtime.py b/tests/providers/test_provider_runtime.py
index 216b8053..f7ac5343 100644
--- a/tests/providers/test_provider_runtime.py
+++ b/tests/providers/test_provider_runtime.py
@@ -25,6 +25,7 @@ from free_claude_code.config.provider_catalog import (
NEBIUS_DEFAULT_BASE,
OLLAMA_CLOUD_DEFAULT_BASE,
PROVIDER_CATALOG,
+ QWENCLOUD_CODING_DEFAULT_BASE,
QWENCLOUD_DEFAULT_BASE,
SILICONFLOW_DEFAULT_BASE,
SUPPORTED_PROVIDER_IDS,
@@ -73,6 +74,7 @@ def _make_settings(**overrides):
mock.open_router_api_key = "test_openrouter_key"
mock.xai_api_key = "test_xai_key"
mock.qwencloud_api_key = "test_qwencloud_key"
+ mock.qwencloud_coding_api_key = "test_qwencloud_coding_key"
mock.together_api_key = "test_together_key"
mock.deepinfra_api_key = "test_deepinfra_key"
mock.siliconflow_api_key = "test_siliconflow_key"
@@ -150,6 +152,7 @@ def _make_settings(**overrides):
mock.openai_proxy = None
mock.xai_proxy = None
mock.qwencloud_proxy = None
+ mock.qwencloud_coding_proxy = None
mock.together_proxy = None
mock.deepinfra_proxy = None
mock.siliconflow_proxy = None
@@ -263,6 +266,25 @@ def test_qwencloud_provider_config_uses_key_base_and_proxy() -> None:
assert isinstance(provider, OpenAIChatProvider)
+def test_qwencloud_coding_provider_config_uses_key_base_and_proxy() -> None:
+ descriptor = PROVIDER_CATALOG["qwencloud_coding"]
+ settings = _make_settings(
+ qwencloud_coding_api_key="qwencloud-coding-token",
+ qwencloud_coding_proxy="http://proxy.test:8080",
+ )
+
+ config = build_provider_config(descriptor, settings)
+ with patch("free_claude_code.providers.openai_chat.provider.AsyncOpenAI"):
+ provider = create_provider("qwencloud_coding", settings)
+
+ assert descriptor.display_name == "QwenCloud Coding Plan"
+ assert descriptor.credential_env == "QWENCLOUD_CODING_API_KEY"
+ assert config.api_key == "qwencloud-coding-token"
+ assert config.base_url == QWENCLOUD_CODING_DEFAULT_BASE
+ assert config.proxy == "http://proxy.test:8080"
+ assert isinstance(provider, OpenAIChatProvider)
+
+
def test_together_provider_config_uses_key_base_and_proxy() -> None:
descriptor = PROVIDER_CATALOG["together"]
settings = _make_settings(
@@ -742,6 +764,7 @@ def test_create_provider_instantiates_each_builtin():
"openai": OpenAICodexProvider,
"xai": OpenAIChatProvider,
"qwencloud": OpenAIChatProvider,
+ "qwencloud_coding": OpenAIChatProvider,
"together": OpenAIChatProvider,
"deepinfra": OpenAIChatProvider,
"siliconflow": OpenAIChatProvider,
diff --git a/tests/providers/test_qwencloud_coding.py b/tests/providers/test_qwencloud_coding.py
new file mode 100644
index 00000000..656c5a6d
--- /dev/null
+++ b/tests/providers/test_qwencloud_coding.py
@@ -0,0 +1,244 @@
+"""Tests for the QwenCloud Coding Plan OpenAI-chat provider profile."""
+
+from types import SimpleNamespace
+
+import httpx
+import pytest
+from openai import AsyncOpenAI
+
+from free_claude_code.application.model_metadata import ProviderModelInfo
+from free_claude_code.config.provider_catalog import QWENCLOUD_CODING_DEFAULT_BASE
+from free_claude_code.core.anthropic.models import MessagesRequest
+from free_claude_code.core.reasoning import ReasoningEffort, ReasoningPolicy
+from free_claude_code.providers.openai_chat import OpenAIChatProvider
+from tests.providers.support import (
+ REASONING_OFF,
+ REASONING_ON,
+ immediate_admission,
+ make_provider_config,
+ profiled_provider,
+ reasoning_for,
+)
+
+
+def qwencloud_coding_provider() -> OpenAIChatProvider:
+ return profiled_provider(
+ "qwencloud_coding",
+ make_provider_config(
+ api_key="test-qwencloud-coding-key",
+ base_url=QWENCLOUD_CODING_DEFAULT_BASE,
+ rate_limit=10,
+ rate_window=60,
+ ),
+ admission=immediate_admission(provider_name="qwencloud_coding"),
+ )
+
+
+def test_init_uses_openai_chat_provider(
+ qwencloud_coding_provider: OpenAIChatProvider,
+) -> None:
+ assert qwencloud_coding_provider._api_key == "test-qwencloud-coding-key"
+ assert qwencloud_coding_provider._base_url == QWENCLOUD_CODING_DEFAULT_BASE
+ assert qwencloud_coding_provider._provider_name == "QWENCLOUD_CODING"
+
+
+def test_build_request_body_preserves_tools_and_images_without_inventing_a_cap(
+ qwencloud_coding_provider: OpenAIChatProvider,
+) -> None:
+ request = MessagesRequest.model_validate(
+ {
+ "model": "qwen3.7-plus",
+ "system": "Be concise.",
+ "messages": [
+ {
+ "role": "user",
+ "content": [
+ {"type": "text", "text": "Inspect this image."},
+ {
+ "type": "image",
+ "source": {
+ "type": "base64",
+ "media_type": "image/png",
+ "data": "aW1hZ2U=",
+ },
+ },
+ ],
+ }
+ ],
+ "tools": [
+ {
+ "name": "inspect_image",
+ "description": "Inspect an image",
+ "input_schema": {
+ "type": "object",
+ "properties": {"detail": {"type": "string"}},
+ },
+ }
+ ],
+ }
+ )
+
+ body = qwencloud_coding_provider._build_request_body(
+ request,
+ reasoning=reasoning_for(request),
+ )
+
+ assert body["model"] == "qwen3.7-plus"
+ assert "max_tokens" not in body
+ assert body["messages"][0] == {"role": "system", "content": "Be concise."}
+ assert body["messages"][1]["content"] == [
+ {"type": "text", "text": "Inspect this image."},
+ {
+ "type": "image_url",
+ "image_url": {"url": "data:image/png;base64,aW1hZ2U="},
+ },
+ ]
+ assert body["tools"][0]["function"]["name"] == "inspect_image"
+ assert "tool_stream" not in body
+
+
+def test_build_request_body_preserves_explicit_client_cap(
+ qwencloud_coding_provider: OpenAIChatProvider,
+) -> None:
+ request = MessagesRequest.model_validate(
+ {
+ "model": "qwen3.7-plus",
+ "max_tokens": 321,
+ "messages": [{"role": "user", "content": "Hello"}],
+ }
+ )
+
+ body = qwencloud_coding_provider._build_request_body(
+ request,
+ reasoning=reasoning_for(request),
+ )
+
+ assert body["max_tokens"] == 321
+
+
+ "reasoning",
+ [
+ REASONING_OFF,
+ REASONING_ON,
+ ReasoningPolicy.on(effort=ReasoningEffort.MAX),
+ ],
+)
+def test_build_request_body_does_not_invent_catalog_wide_reasoning_control(
+ qwencloud_coding_provider: OpenAIChatProvider,
+ reasoning: ReasoningPolicy,
+) -> None:
+ request = MessagesRequest.model_validate(
+ {
+ "model": "qwen3.7-plus",
+ "messages": [{"role": "user", "content": "Hello"}],
+ }
+ )
+
+ body = qwencloud_coding_provider._build_request_body(request, reasoning=reasoning)
+ extra_body = body.get("extra_body", {})
+
+ for field in (
+ "enable_thinking",
+ "preserve_thinking",
+ "reasoning",
+ "reasoning_effort",
+ "thinking",
+ "thinking_budget",
+ "tool_stream",
+ ):
+ assert field not in body
+ assert field not in extra_body
+
+
+def test_build_request_body_replays_prior_reasoning_content(
+ qwencloud_coding_provider: OpenAIChatProvider,
+) -> None:
+ request = MessagesRequest.model_validate(
+ {
+ "model": "qwen3.7-plus",
+ "messages": [
+ {"role": "user", "content": "Solve it."},
+ {
+ "role": "assistant",
+ "content": [
+ {"type": "thinking", "thinking": "Work through it."},
+ {"type": "text", "text": "The answer is 42."},
+ ],
+ },
+ {"role": "user", "content": "Continue."},
+ ],
+ }
+ )
+
+ body = qwencloud_coding_provider._build_request_body(
+ request,
+ reasoning=reasoning_for(request),
+ )
+
+ assert body["messages"][1] == {
+ "role": "assistant",
+ "content": "The answer is 42.",
+ "reasoning_content": "Work through it.",
+ }
+ assert (
+ qwencloud_coding_provider._profile.reasoning_delta(
+ SimpleNamespace(reasoning_content="next thought")
+ )
+ == "next thought"
+ )
+
+
+async def test_model_catalog_uses_standard_endpoint_base_url_and_auth(
+ qwencloud_coding_provider: OpenAIChatProvider,
+) -> None:
+ requests: list[httpx.Request] = []
+
+ def handler(request: httpx.Request) -> httpx.Response:
+ requests.append(request)
+ return httpx.Response(
+ 200,
+ json={
+ "object": "list",
+ "data": [
+ {
+ "id": "qwen3.7-plus",
+ "object": "model",
+ "created": 0,
+ "owned_by": "qwencloud",
+ },
+ {
+ "id": "kimi-k2.5",
+ "object": "model",
+ "created": 0,
+ "owned_by": "qwencloud",
+ },
+ ],
+ },
+ )
+
+ await qwencloud_coding_provider._client.close()
+ qwencloud_coding_provider._client = AsyncOpenAI(
+ api_key="wire-qwencloud-coding-key",
+ base_url=QWENCLOUD_CODING_DEFAULT_BASE,
+ max_retries=0,
+ http_client=httpx.AsyncClient(transport=httpx.MockTransport(handler)),
+ )
+ try:
+ model_infos = await qwencloud_coding_provider.list_model_infos()
+ finally:
+ await qwencloud_coding_provider.cleanup()
+
+ assert model_infos == frozenset(
+ {
+ ProviderModelInfo("qwen3.7-plus"),
+ ProviderModelInfo("kimi-k2.5"),
+ }
+ )
+ assert len(requests) == 1
+ assert str(requests[0].url) == (
+ "https://coding-intl.dashscope.aliyuncs.com/v1/models"
+ )
+ assert requests[0].headers["authorization"] == ("Bearer wire-qwencloud-coding-key")
diff --git a/uv.lock b/uv.lock
index 7934a0e7..effab848 100644
--- a/uv.lock
+++ b/uv.lock
@@ -607,7 +607,7 @@ wheels = [
[[package]]
name = "free-claude-code"
-version = "5.1.2"
+version = "5.2.0"
source = { editable = "." }
dependencies = [
{ name = "aiohttp" },