diff --git a/mateclaw-server/src/main/java/vip/mate/agent/chatmodel/AgentDashScopeChatModelBuilder.java b/mateclaw-server/src/main/java/vip/mate/agent/chatmodel/AgentDashScopeChatModelBuilder.java index 8458a647..4ec000e6 100644 --- a/mateclaw-server/src/main/java/vip/mate/agent/chatmodel/AgentDashScopeChatModelBuilder.java +++ b/mateclaw-server/src/main/java/vip/mate/agent/chatmodel/AgentDashScopeChatModelBuilder.java @@ -70,18 +70,60 @@ public class AgentDashScopeChatModelBuilder implements ChatModelBuilder { } /** - * DashScope's built-in web search is on by default; only an explicit - * {@code enableSearch=false} in provider kwargs disables it. Public so - * {@code AgentGraphBuilder.build()} can surface the "built-in search - * active" log once per agent. + * Bailian's built-in web search is only accepted by a subset of models; + * sending {@code enable_search} to a model that doesn't support it returns + * 400 InvalidParameter and the failover layer then evicts the entire + * provider as MODEL_NOT_FOUND. Per the public docs, only Qwen-Plus, + * Qwen-Max, Qwen-Turbo and the Qwen3-Max series accept the parameter. + * + *

Resolution order:

+ *
    + *
  1. Explicit {@code enableSearch} in the model row (per-model toggle)
    2. + *
  2. Explicit {@code enableSearch} in provider kwargs (admin-level toggle)
    3. + *
  3. Default: enabled only when the model name matches a known-supporting + * prefix; disabled for everything else (coder / thinking / DeepSeek / + * Long / Vision)
    4. + *
+ * + *

Public so {@code AgentGraphBuilder.build()} can surface the + * "built-in search active" log once per agent.

*/ public boolean isBuiltinSearchEnabled(ModelConfigEntity runtimeModel, ModelProviderEntity provider) { + if (runtimeModel != null && runtimeModel.getEnableSearch() != null) { + return Boolean.TRUE.equals(runtimeModel.getEnableSearch()); + } Map kwargs = modelProviderService.readProviderGenerateKwargs(provider); Object kwargsSearch = kwargs.get("enableSearch"); if (kwargsSearch != null) { return Boolean.TRUE.equals(kwargsSearch); } - return true; + return modelSupportsBuiltinSearch(runtimeModel); + } + + /** + * Model id prefixes that the Bailian text-generation endpoint documents as + * accepting {@code enable_search}. Anything else (qwen-coder-*, qwen-long, + * qwen3-*-thinking-*, deepseek-*, qwen*-vl-*) returns 400 InvalidParameter + * when the parameter is sent. + */ + private static final java.util.List BUILTIN_SEARCH_SUPPORTED_PREFIXES = java.util.List.of( + "qwen-plus", + "qwen-max", + "qwen-turbo", + "qwen3-max", + "qwen3.5-flash", + "qwen3.6-flash" + ); + + private static boolean modelSupportsBuiltinSearch(ModelConfigEntity model) { + if (model == null) return false; + String name = model.getModelName(); + if (!StringUtils.hasText(name)) return false; + String lower = name.trim().toLowerCase(); + for (String prefix : BUILTIN_SEARCH_SUPPORTED_PREFIXES) { + if (lower.startsWith(prefix)) return true; + } + return false; } DashScopeChatOptions buildDashScopeOptions(ModelConfigEntity runtimeModel, ModelProviderEntity provider) { diff --git a/mateclaw-server/src/main/resources/db/data-en.sql b/mateclaw-server/src/main/resources/db/data-en.sql index 6757a8f6..27dd7069 100644 --- a/mateclaw-server/src/main/resources/db/data-en.sql +++ b/mateclaw-server/src/main/resources/db/data-en.sql @@ -171,9 +171,13 @@ MERGE INTO mate_model_config (id, name, provider, model_name, description, tempe (1000000101, 'Qwen3 Max', 'dashscope', 'qwen3-max', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000102, 'Qwen3 235B A22B Thinking', 'dashscope', 'qwen3-235b-a22b-thinking-2507', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000103, 'DeepSeek-V3.2', 'dashscope', 'deepseek-v3.2', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), --- Removed: qwen3.5-plus / qwen3.5-max — unavailable on DashScope native protocol (returns 400 InvalidParameter) -(1000000172, 'Qwen3 Plus', 'dashscope', 'qwen3-plus', 'Qwen3 balanced model', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +-- Note: dotted Qwen3 versions (qwen3-plus / qwen3.5-plus / qwen3.5-max / qwen3.6-*) only ship on the +-- OpenAI-compatible endpoint. Calling them through DashScope native (text-generation/generation) +-- returns 400 InvalidParameter — use the bailian-team OpenAI-compat provider instead. (1000000173, 'Qwen Long', 'dashscope', 'qwen-long', 'Long-context model with extended context support', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000174, 'Qwen Plus (latest)', 'dashscope', 'qwen-plus-latest', 'Latest stable snapshot of Qwen Plus — auto-updates as Bailian rolls new releases', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000175, 'Qwen Max (latest)', 'dashscope', 'qwen-max-latest', 'Latest stable snapshot of Qwen Max — strongest reasoning capability', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000176, 'Qwen Turbo (latest)', 'dashscope', 'qwen-turbo-latest', 'Latest stable snapshot of Qwen Turbo — low latency, high frequency', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000104, 'Qwen3.5-122B-A10B', 'modelscope', 'Qwen/Qwen3.5-122B-A10B', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000105, 'GLM-5', 'modelscope', 'ZhipuAI/GLM-5', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000106, 'Qwen3.5 Plus', 'aliyun-codingplan', 'qwen3.5-plus', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), diff --git a/mateclaw-server/src/main/resources/db/data-mysql-en.sql b/mateclaw-server/src/main/resources/db/data-mysql-en.sql index 5a82aa31..556892e4 100644 --- a/mateclaw-server/src/main/resources/db/data-mysql-en.sql +++ b/mateclaw-server/src/main/resources/db/data-mysql-en.sql @@ -186,9 +186,13 @@ VALUES (1000000101, 'Qwen3 Max', 'dashscope', 'qwen3-max', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000102, 'Qwen3 235B A22B Thinking', 'dashscope', 'qwen3-235b-a22b-thinking-2507', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000103, 'DeepSeek-V3.2', 'dashscope', 'deepseek-v3.2', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), --- Removed: qwen3.5-plus / qwen3.5-max — unavailable on DashScope native protocol (returns 400 InvalidParameter) -(1000000172, 'Qwen3 Plus', 'dashscope', 'qwen3-plus', 'Qwen3 balanced model', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +-- Note: dotted Qwen3 versions (qwen3-plus / qwen3.5-plus / qwen3.5-max / qwen3.6-*) only ship on the +-- OpenAI-compatible endpoint. Calling them through DashScope native (text-generation/generation) +-- returns 400 InvalidParameter — use the bailian-team OpenAI-compat provider instead. (1000000173, 'Qwen Long', 'dashscope', 'qwen-long', 'Long-context model with extended context support', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000174, 'Qwen Plus (latest)', 'dashscope', 'qwen-plus-latest', 'Latest stable snapshot of Qwen Plus — auto-updates as Bailian rolls new releases', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000175, 'Qwen Max (latest)', 'dashscope', 'qwen-max-latest', 'Latest stable snapshot of Qwen Max — strongest reasoning capability', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000176, 'Qwen Turbo (latest)', 'dashscope', 'qwen-turbo-latest', 'Latest stable snapshot of Qwen Turbo — low latency, high frequency', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000104, 'Qwen3.5-122B-A10B', 'modelscope', 'Qwen/Qwen3.5-122B-A10B', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000105, 'GLM-5', 'modelscope', 'ZhipuAI/GLM-5', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000106, 'Qwen3.5 Plus', 'aliyun-codingplan', 'qwen3.5-plus', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), @@ -206,6 +210,16 @@ VALUES (1000000404, 'Qwen Image 2.0 Pro', 'bailian-team', 'qwen-image-2.0-pro', 'Bailian Token Plan — Qwen image generation flagship model', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000405, 'Wan 2.7 Image', 'bailian-team', 'wan2.7-image', 'Bailian Token Plan — Wan image generation model', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000406, 'Wan 2.7 Image Pro', 'bailian-team', 'wan2.7-image-pro', 'Bailian Token Plan — Wan image generation flagship model', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000407, 'Qwen 3.5 Plus', 'bailian-team', 'qwen3.5-plus', 'Bailian Token Plan — Qwen3.5 balanced flagship, hybrid thinking, 128K context', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000408, 'Qwen 3.5 Flash', 'bailian-team', 'qwen3.5-flash', 'Bailian Token Plan — Qwen3.5 fast variant for high-frequency calls', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000409, 'Qwen3 VL Plus', 'bailian-team', 'qwen3-vl-plus', 'Bailian Token Plan — Qwen3 vision-language flagship, image + video', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000410, 'Qwen3 VL Flash', 'bailian-team', 'qwen3-vl-flash', 'Bailian Token Plan — Qwen3 vision-language fast variant', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000411, 'Qwen3 Coder Plus', 'bailian-team', 'qwen3-coder-plus', 'Bailian Token Plan — Qwen3 coding flagship, agentic code editing & tools', 0.2, 8192, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000412, 'Qwen 3.6 Plus 2026-04-02', 'bailian-team', 'qwen3.6-plus-2026-04-02', 'Bailian Token Plan — pinned snapshot of Qwen 3.6 Plus released 2026-04-02', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000413, 'Qwen 3.6 Max (preview)', 'bailian-team', 'qwen3.6-max-preview', 'Bailian Token Plan — Qwen3.6 Max preview, strongest 3.6 reasoning', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000414, 'Qwen 3.6 Flash', 'bailian-team', 'qwen3.6-flash', 'Bailian Token Plan — Qwen3.6 fast variant, hybrid thinking default-on', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000415, 'Qwen 3.6 Flash 2026-04-16','bailian-team', 'qwen3.6-flash-2026-04-16','Bailian Token Plan — pinned snapshot of Qwen 3.6 Flash released 2026-04-16', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000416, 'Qwen 3.5 Omni Plus', 'bailian-team', 'qwen3.5-omni-plus', 'Bailian Token Plan — Qwen3.5 omni-modal plus, text + vision + audio in/out', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000114, 'GPT-5.2', 'openai', 'gpt-5.2', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000115, 'GPT-5', 'openai', 'gpt-5', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000116, 'GPT-5 Mini', 'openai', 'gpt-5-mini', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), diff --git a/mateclaw-server/src/main/resources/db/data-mysql-zh.sql b/mateclaw-server/src/main/resources/db/data-mysql-zh.sql index 33f6ecaf..7307209c 100644 --- a/mateclaw-server/src/main/resources/db/data-mysql-zh.sql +++ b/mateclaw-server/src/main/resources/db/data-mysql-zh.sql @@ -186,9 +186,12 @@ VALUES (1000000101, 'Qwen3 Max', 'dashscope', 'qwen3-max', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000102, 'Qwen3 235B A22B Thinking', 'dashscope', 'qwen3-235b-a22b-thinking-2507', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000103, 'DeepSeek-V3.2', 'dashscope', 'deepseek-v3.2', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), --- Removed: qwen3.5-plus / qwen3.5-max — unavailable on DashScope native protocol (returns 400 InvalidParameter) -(1000000172, 'Qwen3 Plus', 'dashscope', 'qwen3-plus', 'Qwen3 均衡模型', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +-- 注意: qwen3-plus / qwen3.5-plus / qwen3.5-max / qwen3.6-* 等带点号的版本只在 OpenAI 兼容端点上线， +-- DashScope native（text-generation/generation）调用会返回 400 InvalidParameter，请使用 bailian-team 等 OpenAI-compat provider。 (1000000173, 'Qwen Long', 'dashscope', 'qwen-long', '长文本模型，支持超长上下文', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000174, 'Qwen Plus (latest)', 'dashscope', 'qwen-plus-latest', '通义千问 Plus 最新稳定快照，自动跟随官方更新', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000175, 'Qwen Max (latest)', 'dashscope', 'qwen-max-latest', '通义千问 Max 最新稳定快照，最强推理能力', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000176, 'Qwen Turbo (latest)', 'dashscope', 'qwen-turbo-latest', '通义千问 Turbo 最新稳定快照，低延迟、高并发', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000104, 'Qwen3.5-122B-A10B', 'modelscope', 'Qwen/Qwen3.5-122B-A10B', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000105, 'GLM-5', 'modelscope', 'ZhipuAI/GLM-5', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000106, 'Qwen3.5 Plus', 'aliyun-codingplan', 'qwen3.5-plus', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), @@ -206,6 +209,16 @@ VALUES (1000000404, 'Qwen Image 2.0 Pro', 'bailian-team', 'qwen-image-2.0-pro', '百炼团队套餐 — 千问图片生成旗舰模型', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000405, 'Wan 2.7 Image', 'bailian-team', 'wan2.7-image', '百炼团队套餐 — 万相图片生成模型', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000406, 'Wan 2.7 Image Pro', 'bailian-team', 'wan2.7-image-pro', '百炼团队套餐 — 万相图片生成旗舰模型', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000407, 'Qwen 3.5 Plus', 'bailian-team', 'qwen3.5-plus', '百炼团队套餐 — Qwen3.5 均衡旗舰，混合思考，128K 上下文', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000408, 'Qwen 3.5 Flash', 'bailian-team', 'qwen3.5-flash', '百炼团队套餐 — Qwen3.5 快速版，低延迟、高并发', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000409, 'Qwen3 VL Plus', 'bailian-team', 'qwen3-vl-plus', '百炼团队套餐 — Qwen3 视觉旗舰，支持图像与视频理解', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000410, 'Qwen3 VL Flash', 'bailian-team', 'qwen3-vl-flash', '百炼团队套餐 — Qwen3 视觉快速版，高吞吐视觉调用', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000411, 'Qwen3 Coder Plus', 'bailian-team', 'qwen3-coder-plus', '百炼团队套餐 — Qwen3 编码旗舰，智能体代码编辑与工具调用', 0.2, 8192, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000412, 'Qwen 3.6 Plus 2026-04-02', 'bailian-team', 'qwen3.6-plus-2026-04-02', '百炼团队套餐 — Qwen 3.6 Plus 2026-04-02 锁定快照', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000413, 'Qwen 3.6 Max (preview)', 'bailian-team', 'qwen3.6-max-preview', '百炼团队套餐 — Qwen3.6 Max 预览版，3.6 系列最强推理', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000414, 'Qwen 3.6 Flash', 'bailian-team', 'qwen3.6-flash', '百炼团队套餐 — Qwen3.6 快速版，混合思考默认开启', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000415, 'Qwen 3.6 Flash 2026-04-16','bailian-team', 'qwen3.6-flash-2026-04-16','百炼团队套餐 — Qwen 3.6 Flash 2026-04-16 锁定快照', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000416, 'Qwen 3.5 Omni Plus', 'bailian-team', 'qwen3.5-omni-plus', '百炼团队套餐 — Qwen3.5 全模态版，文本/视觉/音频输入输出', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000114, 'GPT-5.2', 'openai', 'gpt-5.2', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000115, 'GPT-5', 'openai', 'gpt-5', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000116, 'GPT-5 Mini', 'openai', 'gpt-5-mini', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), diff --git a/mateclaw-server/src/main/resources/db/data-zh.sql b/mateclaw-server/src/main/resources/db/data-zh.sql index 55b8234d..b0859a41 100644 --- a/mateclaw-server/src/main/resources/db/data-zh.sql +++ b/mateclaw-server/src/main/resources/db/data-zh.sql @@ -175,9 +175,12 @@ MERGE INTO mate_model_config (id, name, provider, model_name, description, tempe (1000000101, 'Qwen3 Max', 'dashscope', 'qwen3-max', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000102, 'Qwen3 235B A22B Thinking', 'dashscope', 'qwen3-235b-a22b-thinking-2507', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000103, 'DeepSeek-V3.2', 'dashscope', 'deepseek-v3.2', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), --- Removed: qwen3.5-plus / qwen3.5-max — unavailable on DashScope native protocol (returns 400 InvalidParameter) -(1000000172, 'Qwen3 Plus', 'dashscope', 'qwen3-plus', 'Qwen3 均衡模型', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +-- 注意: qwen3-plus / qwen3.5-plus / qwen3.5-max / qwen3.6-* 等带点号的版本只在 OpenAI 兼容端点上线， +-- DashScope native（text-generation/generation）调用会返回 400 InvalidParameter，请使用 bailian-team 等 OpenAI-compat provider。 (1000000173, 'Qwen Long', 'dashscope', 'qwen-long', '长文本模型，支持超长上下文', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000174, 'Qwen Plus (latest)', 'dashscope', 'qwen-plus-latest', '通义千问 Plus 最新稳定快照，自动跟随官方更新', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000175, 'Qwen Max (latest)', 'dashscope', 'qwen-max-latest', '通义千问 Max 最新稳定快照，最强推理能力', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), +(1000000176, 'Qwen Turbo (latest)', 'dashscope', 'qwen-turbo-latest', '通义千问 Turbo 最新稳定快照，低延迟、高并发', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000104, 'Qwen3.5-122B-A10B', 'modelscope', 'Qwen/Qwen3.5-122B-A10B', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000105, 'GLM-5', 'modelscope', 'ZhipuAI/GLM-5', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), (1000000106, 'Qwen3.5 Plus', 'aliyun-codingplan', 'qwen3.5-plus', '', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, NOW(), NOW(), 0), diff --git a/mateclaw-server/src/main/resources/db/migration/h2/V81__refresh_bailian_qwen_catalog.sql b/mateclaw-server/src/main/resources/db/migration/h2/V81__refresh_bailian_qwen_catalog.sql new file mode 100644 index 00000000..67d5e2ae --- /dev/null +++ b/mateclaw-server/src/main/resources/db/migration/h2/V81__refresh_bailian_qwen_catalog.sql @@ -0,0 +1,85 @@ +-- Refresh the Bailian (Aliyun Model Studio) Qwen catalog to match the 2026 Q2 +-- model lineup, and remove a non-existent model id that triggered 400 +-- InvalidParameter on the native text-generation endpoint. +-- +-- Why now: +-- * Reported user error: "Bad request, please check input (type=MODEL_NOT_FOUND)" +-- when chatting with the seeded "Qwen3 Plus" entry. +-- * Root cause: model id `qwen3-plus` does not exist on Bailian. The real +-- balanced Qwen3 series uses dotted minor versions (`qwen3.5-plus`, +-- `qwen3.6-plus`); plain `qwen3-plus` was never published. DashScope's +-- native endpoint rejects it with `[InvalidParameter]`, which our +-- failover classifier maps to MODEL_NOT_FOUND and evicts the entire +-- dashscope provider from the pool. +-- +-- Two-part fix: +-- 1. Soft-delete the bogus `qwen3-plus` row (id 1000000172). +-- 2. Seed three latest-snapshot trackers on the dashscope native provider +-- (qwen-plus-latest / qwen-max-latest / qwen-turbo-latest) and six +-- newer Qwen3 series models on the bailian-team OpenAI-compat provider +-- where they are documented to work (vision + flash + coder + 3.6 snapshot). + +-- 1. Soft-delete the bogus model id (idempotent). +UPDATE mate_model_config + SET deleted = 1, enabled = FALSE, update_time = CURRENT_TIMESTAMP + WHERE id = 1000000172 + AND model_name = 'qwen3-plus'; + +-- 2a. Latest-snapshot trackers on dashscope native (text-generation endpoint). +-- These ids forward to whatever Bailian currently considers the stable +-- release of the series, so users don't have to chase dated snapshots. +MERGE INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +KEY (id) +VALUES (1000000174, 'Qwen Plus (latest)', 'dashscope', 'qwen-plus-latest', 'Latest stable snapshot of Qwen Plus — auto-updates as Bailian rolls new releases.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 0); + +MERGE INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +KEY (id) +VALUES (1000000175, 'Qwen Max (latest)', 'dashscope', 'qwen-max-latest', 'Latest stable snapshot of Qwen Max — strongest reasoning capability.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 0); + +MERGE INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +KEY (id) +VALUES (1000000176, 'Qwen Turbo (latest)', 'dashscope', 'qwen-turbo-latest', 'Latest stable snapshot of Qwen Turbo — low latency, high frequency.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 0); + +-- 2b. Newer Qwen3 series on the bailian-team OpenAI-compat token plan endpoint. +-- These models target the OpenAI-compatible URL and accept the same +-- tool-call schema as any other OpenAI-compat provider. + +MERGE INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +KEY (id) +VALUES (1000000407, 'Qwen 3.5 Plus', 'bailian-team', 'qwen3.5-plus', 'Bailian Token Plan — Qwen3.5 balanced flagship, hybrid thinking, 128K context.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 0); + +MERGE INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +KEY (id) +VALUES (1000000408, 'Qwen 3.5 Flash', 'bailian-team', 'qwen3.5-flash', 'Bailian Token Plan — Qwen3.5 fast variant, lower latency for high-frequency calls.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 0); + +MERGE INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +KEY (id) +VALUES (1000000409, 'Qwen3 VL Plus', 'bailian-team', 'qwen3-vl-plus', 'Bailian Token Plan — Qwen3 vision-language flagship, image + video understanding.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 0); + +MERGE INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +KEY (id) +VALUES (1000000410, 'Qwen3 VL Flash', 'bailian-team', 'qwen3-vl-flash', 'Bailian Token Plan — Qwen3 vision-language fast variant for high-throughput vision.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 0); + +MERGE INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +KEY (id) +VALUES (1000000411, 'Qwen3 Coder Plus', 'bailian-team', 'qwen3-coder-plus', 'Bailian Token Plan — Qwen3 coding flagship, agentic code editing and tool use.', 0.2, 8192, 0.8, TRUE, TRUE, FALSE, 'chat', CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 0); + +MERGE INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +KEY (id) +VALUES (1000000412, 'Qwen 3.6 Plus 2026-04-02', 'bailian-team', 'qwen3.6-plus-2026-04-02', 'Bailian Token Plan — pinned snapshot of Qwen 3.6 Plus released 2026-04-02.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 0); + +MERGE INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +KEY (id) +VALUES (1000000413, 'Qwen 3.6 Max (preview)', 'bailian-team', 'qwen3.6-max-preview', 'Bailian Token Plan — Qwen3.6 Max preview, strongest reasoning in the 3.6 lineup.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 0); + +MERGE INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +KEY (id) +VALUES (1000000414, 'Qwen 3.6 Flash', 'bailian-team', 'qwen3.6-flash', 'Bailian Token Plan — Qwen3.6 fast variant, hybrid thinking mode default-on.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 0); + +MERGE INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +KEY (id) +VALUES (1000000415, 'Qwen 3.6 Flash 2026-04-16', 'bailian-team', 'qwen3.6-flash-2026-04-16', 'Bailian Token Plan — pinned snapshot of Qwen 3.6 Flash released 2026-04-16.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 0); + +MERGE INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +KEY (id) +VALUES (1000000416, 'Qwen 3.5 Omni Plus', 'bailian-team', 'qwen3.5-omni-plus', 'Bailian Token Plan — Qwen3.5 omni-modal plus, text + vision + audio in/out.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', CURRENT_TIMESTAMP, CURRENT_TIMESTAMP, 0); diff --git a/mateclaw-server/src/main/resources/db/migration/mysql/V81__refresh_bailian_qwen_catalog.sql b/mateclaw-server/src/main/resources/db/migration/mysql/V81__refresh_bailian_qwen_catalog.sql new file mode 100644 index 00000000..d0861b6e --- /dev/null +++ b/mateclaw-server/src/main/resources/db/migration/mysql/V81__refresh_bailian_qwen_catalog.sql @@ -0,0 +1,49 @@ +-- Refresh the Bailian (Aliyun Model Studio) Qwen catalog to match the 2026 Q2 +-- model lineup, and remove a non-existent model id that triggered 400 +-- InvalidParameter on the native text-generation endpoint. +-- +-- Why now: +-- * Reported user error: "Bad request, please check input (type=MODEL_NOT_FOUND)" +-- when chatting with the seeded "Qwen3 Plus" entry. +-- * Root cause: model id `qwen3-plus` does not exist on Bailian. The real +-- balanced Qwen3 series uses dotted minor versions (`qwen3.5-plus`, +-- `qwen3.6-plus`); plain `qwen3-plus` was never published. DashScope's +-- native endpoint rejects it with `[InvalidParameter]`, which our +-- failover classifier maps to MODEL_NOT_FOUND and evicts the entire +-- dashscope provider from the pool. +-- +-- Two-part fix: +-- 1. Soft-delete the bogus `qwen3-plus` row (id 1000000172). +-- 2. Seed three latest-snapshot trackers on the dashscope native provider +-- (qwen-plus-latest / qwen-max-latest / qwen-turbo-latest) and six +-- newer Qwen3 series models on the bailian-team OpenAI-compat provider +-- where they are documented to work (vision + flash + coder + 3.6 snapshot). + +-- 1. Soft-delete the bogus model id (idempotent). +UPDATE mate_model_config + SET deleted = 1, enabled = FALSE, update_time = NOW() + WHERE id = 1000000172 + AND model_name = 'qwen3-plus'; + +-- 2a. Latest-snapshot trackers on dashscope native (text-generation endpoint). +INSERT INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +VALUES +(1000000174, 'Qwen Plus (latest)', 'dashscope', 'qwen-plus-latest', 'Latest stable snapshot of Qwen Plus — auto-updates as Bailian rolls new releases.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', NOW(), NOW(), 0), +(1000000175, 'Qwen Max (latest)', 'dashscope', 'qwen-max-latest', 'Latest stable snapshot of Qwen Max — strongest reasoning capability.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', NOW(), NOW(), 0), +(1000000176, 'Qwen Turbo (latest)', 'dashscope', 'qwen-turbo-latest', 'Latest stable snapshot of Qwen Turbo — low latency, high frequency.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', NOW(), NOW(), 0) +ON DUPLICATE KEY UPDATE name=VALUES(name), description=VALUES(description), model_type=VALUES(model_type), update_time=VALUES(update_time); + +-- 2b. Newer Qwen3 series on the bailian-team OpenAI-compat token plan endpoint. +INSERT INTO mate_model_config (id, name, provider, model_name, description, temperature, max_tokens, top_p, builtin, enabled, is_default, model_type, create_time, update_time, deleted) +VALUES +(1000000407, 'Qwen 3.5 Plus', 'bailian-team', 'qwen3.5-plus', 'Bailian Token Plan — Qwen3.5 balanced flagship, hybrid thinking, 128K context.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', NOW(), NOW(), 0), +(1000000408, 'Qwen 3.5 Flash', 'bailian-team', 'qwen3.5-flash', 'Bailian Token Plan — Qwen3.5 fast variant, lower latency for high-frequency calls.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', NOW(), NOW(), 0), +(1000000409, 'Qwen3 VL Plus', 'bailian-team', 'qwen3-vl-plus', 'Bailian Token Plan — Qwen3 vision-language flagship, image + video understanding.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', NOW(), NOW(), 0), +(1000000410, 'Qwen3 VL Flash', 'bailian-team', 'qwen3-vl-flash', 'Bailian Token Plan — Qwen3 vision-language fast variant for high-throughput vision.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', NOW(), NOW(), 0), +(1000000411, 'Qwen3 Coder Plus', 'bailian-team', 'qwen3-coder-plus', 'Bailian Token Plan — Qwen3 coding flagship, agentic code editing and tool use.', 0.2, 8192, 0.8, TRUE, TRUE, FALSE, 'chat', NOW(), NOW(), 0), +(1000000412, 'Qwen 3.6 Plus 2026-04-02', 'bailian-team', 'qwen3.6-plus-2026-04-02','Bailian Token Plan — pinned snapshot of Qwen 3.6 Plus released 2026-04-02.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', NOW(), NOW(), 0), +(1000000413, 'Qwen 3.6 Max (preview)', 'bailian-team', 'qwen3.6-max-preview', 'Bailian Token Plan — Qwen3.6 Max preview, strongest reasoning in the 3.6 lineup.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', NOW(), NOW(), 0), +(1000000414, 'Qwen 3.6 Flash', 'bailian-team', 'qwen3.6-flash', 'Bailian Token Plan — Qwen3.6 fast variant, hybrid thinking mode default-on.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', NOW(), NOW(), 0), +(1000000415, 'Qwen 3.6 Flash 2026-04-16','bailian-team', 'qwen3.6-flash-2026-04-16','Bailian Token Plan — pinned snapshot of Qwen 3.6 Flash released 2026-04-16.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', NOW(), NOW(), 0), +(1000000416, 'Qwen 3.5 Omni Plus', 'bailian-team', 'qwen3.5-omni-plus', 'Bailian Token Plan — Qwen3.5 omni-modal plus, text + vision + audio in/out.', 0.7, 4096, 0.8, TRUE, TRUE, FALSE, 'chat', NOW(), NOW(), 0) +ON DUPLICATE KEY UPDATE name=VALUES(name), description=VALUES(description), model_type=VALUES(model_type), update_time=VALUES(update_time);