BCAI/easyai-ai-gateway
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

feat(models): 注册 Kimi K3 和 Qwen3.8

Browse Source
This commit is contained in:
wangbo
2026-07-27 23:50:17 +08:00
parent 31c32690b2
commit 039220c835
3 changed files with 478 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
apps/api/internal/clients/chat_reasoning.go
+24
View File
@@ -78,6 +78,15 @@ func applyAliyunReasoning(body map[string]any, candidate store.RuntimeModelCandi
body["enable_thinking"] = true
model := chatReasoningModelName(body, candidate)
if isAliyunQwen38MaxPreview(model) {
if budget, ok := positiveIntFromAny(body["thinking_budget_tokens"]); ok {
body["thinking_budget"] = budget
delete(body, "reasoning_effort")
return
}
body["reasoning_effort"] = qwen38MaxPreviewReasoningEffort(effort)
return
}
if isAliyunHighMaxReasoningModel(model) {
body["reasoning_effort"] = highMaxReasoningEffort(effort)
return
@@ -200,6 +209,10 @@ func isAliyunHighMaxReasoningModel(model string) bool {
return strings.Contains(model, "deepseek-v4") || strings.HasPrefix(model, "glm-")
}
func isAliyunQwen38MaxPreview(model string) bool {
return strings.Contains(model, "qwen3.8-max-preview")
}
func isAliyunThinkingBudgetModel(model string) bool {
return strings.Contains(model, "qwen") || strings.Contains(model, "qwq") || strings.Contains(model, "qvq") || strings.Contains(model, "kimi")
}
@@ -219,6 +232,17 @@ func highMaxReasoningEffort(effort string) string {
return "high"
}
func qwen38MaxPreviewReasoningEffort(effort string) string {
switch effort {
case "low", "minimal":
return "low"
case "medium":
return "medium"
default:
return "xhigh"
}
}
func zhipuReasoningEffort(effort string) string {
if effort == "max" {
return "xhigh"
apps/api/internal/clients/clients_test.go
+30
View File
@@ -529,6 +529,36 @@ func TestOpenAIClientChatReasoningParamsByProvider(t *testing.T) {
}
},
},
{
name: "aliyun qwen 3.8 preview preserves supported effort",
provider: "aliyun-bailian-openai",
model: "qwen3.8-max-preview",
body: map[string]any{
"reasoning_effort": "high",
},
assertion: func(t *testing.T, body map[string]any) {
if body["enable_thinking"] != true || body["reasoning_effort"] != "xhigh" {
t.Fatalf("aliyun qwen 3.8 should use xhigh reasoning effort, got %+v", body)
}
},
},
{
name: "aliyun qwen 3.8 preview uses explicit budget exclusively",
provider: "aliyun-bailian-openai",
model: "qwen3.8-max-preview",
body: map[string]any{
"reasoning_effort": "medium",
"thinking_budget_tokens": 16384,
},
assertion: func(t *testing.T, body map[string]any) {
if body["enable_thinking"] != true || body["thinking_budget"] != float64(16384) {
t.Fatalf("aliyun qwen 3.8 should use the explicit thinking budget, got %+v", body)
}
if _, ok := body["reasoning_effort"]; ok {
t.Fatalf("aliyun qwen 3.8 must not combine reasoning effort with thinking budget: %+v", body)
}
},
},
{
name: "same deepseek model on aliyun uses aliyun protocol",
provider: "aliyun-bailian-openai",
apps/api/migrations/0088_kimi_k3_qwen38_base_models.sql
+424
View File
@@ -0,0 +1,424 @@
-- Register the latest Kimi and Qwen foundation models in both the EasyAI
-- catalog and their real upstream providers.
--
-- Official references:
-- https://www.volcengine.com/docs/82379/2578669
-- https://help.aliyun.com/zh/model-studio/token-plan-personal-overview
-- https://help.aliyun.com/zh/model-studio/qwen-function-calling
-- https://help.aliyun.com/zh/model-studio/qwen-api-via-openai-responses
WITH model_defs AS (
SELECT *
FROM (
VALUES
(
'easyai',
'easyai',
'EasyAI',
'easyai',
'easyai:Kimi-K3',
'Kimi-K3',
'Kimi-K3',
'Kimi-K3',
'Kimi-K3',
'Kimi 最新旗舰基础模型，原生支持视觉理解，面向 Coding 与 Agent 场景，统一模型 ID 持续升级，支持 100 万上下文。',
'https://static.51easyai.com/kimi-logo-124.png',
'["text_generate","image_analysis","tools_call"]'::jsonb,
'{
"text_generate": {
"supportedApiProtocols": ["openai_chat_completions", "openai_responses"],
"max_context_tokens": 1000000,
"supportTool": true,
"supportThinking": true
},
"image_analysis": {
"max_context_tokens": 1000000,
"supportThinking": true
},
"tools_call": {
"supportedApiProtocols": ["openai_chat_completions", "openai_responses"],
"max_context_tokens": 1000000,
"supportTool": true,
"supportThinking": true
},
"originalTypes": ["text_generate", "image_analysis", "tools_call"]
}'::jsonb,
'easyai:Kimi-K2.7-Code',
'https://www.volcengine.com/docs/82379/2578669',
false
),
(
'volces-openai',
'volces-openai',
'火山引擎(OpenAI兼容)',
'openai',
'volces-openai:kimi-k3',
'kimi-k3',
'kimi-k3',
'Kimi-K3',
'Kimi-K3',
'Kimi 最新旗舰基础模型，原生支持视觉理解，面向 Coding 与 Agent 场景，统一模型 ID 持续升级，支持 100 万上下文。',
'https://static.51easyai.com/kimi-logo-124.png',
'["text_generate","image_analysis","tools_call"]'::jsonb,
'{
"text_generate": {
"supportedApiProtocols": ["openai_chat_completions", "openai_responses"],
"max_context_tokens": 1000000,
"supportTool": true,
"supportThinking": true
},
"image_analysis": {
"max_context_tokens": 1000000,
"supportThinking": true
},
"tools_call": {
"supportedApiProtocols": ["openai_chat_completions", "openai_responses"],
"max_context_tokens": 1000000,
"supportTool": true,
"supportThinking": true
},
"originalTypes": ["text_generate", "image_analysis", "tools_call"]
}'::jsonb,
'volces-openai:doubao-seed-2-0-pro-260215',
'https://www.volcengine.com/docs/82379/2578669',
false
),
(
'easyai',
'easyai',
'EasyAI',
'easyai',
'easyai:Qwen3.8-Max-Preview',
'Qwen3.8-Max-Preview',
'Qwen3.8-Max-Preview',
'Qwen3.8-Max-Preview',
'Qwen3.8-Max-Preview',
'通义千问最新一代 2.4T 旗舰基础模型预览版，支持推理、视觉理解、文本生成与工具调用；仅面向阿里云百炼 Token Plan，需使用订阅专属 Base URL 与 API Key。',
'https://static.51easyai.com/qwen-color.webp',
'["text_generate","image_analysis","tools_call"]'::jsonb,
'{
"text_generate": {
"supportedApiProtocols": ["openai_chat_completions", "openai_responses"],
"supportTool": true,
"supportThinking": true,
"supportThinkingModeSwitch": true,
"thinkingEffortLevels": ["none", "minimal", "low", "medium", "high", "xhigh", "max"],
"thinkingBudgetTokensRange": [4096, 262144],
"supportWebSearch": true
},
"image_analysis": {
"supportThinking": true,
"supportThinkingModeSwitch": true,
"thinkingEffortLevels": ["none", "minimal", "low", "medium", "high", "xhigh", "max"],
"thinkingBudgetTokensRange": [4096, 262144]
},
"tools_call": {
"supportedApiProtocols": ["openai_chat_completions", "openai_responses"],
"supportTool": true,
"supportThinking": true,
"supportThinkingModeSwitch": true,
"thinkingEffortLevels": ["none", "minimal", "low", "medium", "high", "xhigh", "max"],
"thinkingBudgetTokensRange": [4096, 262144],
"supportWebSearch": true
},
"originalTypes": ["text_generate", "image_analysis", "tools_call"]
}'::jsonb,
'easyai:Qwen3.7-Max',
'https://help.aliyun.com/zh/model-studio/token-plan-personal-overview',
true
),
(
'aliyun-bailian-openai',
'aliyun-bailian-openai',
'阿里云百炼(OpenAI兼容）',
'openai',
'aliyun-bailian-openai:qwen3.8-max-preview',
'qwen3.8-max-preview',
'qwen3.8-max-preview',
'Qwen3.8-Max-Preview',
'Qwen3.8-Max-Preview',
'通义千问最新一代 2.4T 旗舰基础模型预览版，支持推理、视觉理解、文本生成与工具调用；仅面向阿里云百炼 Token Plan，需使用订阅专属 Base URL 与 API Key。',
'https://static.51easyai.com/qwen-color.webp',
'["text_generate","image_analysis","tools_call"]'::jsonb,
'{
"text_generate": {
"supportedApiProtocols": ["openai_chat_completions", "openai_responses"],
"supportTool": true,
"supportThinking": true,
"supportThinkingModeSwitch": true,
"thinkingEffortLevels": ["none", "minimal", "low", "medium", "high", "xhigh", "max"],
"thinkingBudgetTokensRange": [4096, 262144],
"supportWebSearch": true
},
"image_analysis": {
"supportThinking": true,
"supportThinkingModeSwitch": true,
"thinkingEffortLevels": ["none", "minimal", "low", "medium", "high", "xhigh", "max"],
"thinkingBudgetTokensRange": [4096, 262144]
},
"tools_call": {
"supportedApiProtocols": ["openai_chat_completions", "openai_responses"],
"supportTool": true,
"supportThinking": true,
"supportThinkingModeSwitch": true,
"thinkingEffortLevels": ["none", "minimal", "low", "medium", "high", "xhigh", "max"],
"thinkingBudgetTokensRange": [4096, 262144],
"supportWebSearch": true
},
"originalTypes": ["text_generate", "image_analysis", "tools_call"]
}'::jsonb,
'aliyun-bailian-openai:qwen3.7-max',
'https://help.aliyun.com/zh/model-studio/token-plan-personal-overview',
true
)
) AS defs(
provider_key,
source_provider_code,
source_provider_name,
source_spec_type,
canonical_model_key,
provider_model_name,
raw_model_name,
display_name,
model_alias,
description,
icon_path,
model_type,
capabilities,
template_key,
reference_document,
token_plan_only
)
),
source_rows AS (
SELECT
providers.id AS provider_id,
model_defs.*,
COALESCE(
template.base_billing_config,
'{"text":{"basePrice":0.01,"baseWeight":1},"image":{"basePrice":10,"baseWeight":1},"currency":"resource"}'::jsonb
) AS base_billing_config,
COALESCE(
template.default_rate_limit_policy,
jsonb_build_object(
'platformLimits',
jsonb_build_object('max_concurrent_requests', 5)
)
) AS default_rate_limit_policy,
COALESCE(
template.pricing_rule_set_id,
(SELECT id FROM model_pricing_rule_sets WHERE rule_set_key = 'default-multimodal-v1' LIMIT 1)
) AS pricing_rule_set_id,
COALESCE(
template.runtime_policy_set_id,
(SELECT id FROM model_runtime_policy_sets WHERE policy_key = 'default-runtime-v1' LIMIT 1)
) AS runtime_policy_set_id,
COALESCE(template.runtime_policy_override, '{}'::jsonb) AS runtime_policy_override,
COALESCE(template.pricing_version, 1) AS pricing_version
FROM model_defs
LEFT JOIN model_catalog_providers providers
ON providers.provider_key = model_defs.provider_key
OR providers.provider_code = model_defs.provider_key
LEFT JOIN base_model_catalog template
ON template.canonical_model_key = model_defs.template_key
),
payload AS (
SELECT
source_rows.*,
jsonb_build_object(
'source', 'server-main.integration-platform',
'sourceProviderCode', source_provider_code,
'sourceProviderName', source_provider_name,
'sourceSpecType', source_spec_type,
'officialModelId', provider_model_name,
'officialDisplayName', display_name,
'originalTypes', model_type,
'alias', model_alias,
'description', description,
'iconPath', icon_path,
'billingType', 'external-api',
'billingMode', '',
'referenceModel', '',
'modelWeight', NULL,
'selectable', true,
'referenceDocument', reference_document,
'tokenPlanOnly', token_plan_only,
'rawModel', jsonb_build_object(
'name', raw_model_name,
'types', model_type,
'icon_path', icon_path,
'alias', model_alias,
'description', description,
'capabilities', capabilities - 'originalTypes'
)
) AS metadata
FROM source_rows
),
snapshot AS (
SELECT
payload.*,
jsonb_build_object(
'providerKey', provider_key,
'canonicalModelKey', canonical_model_key,
'invocationName', provider_model_name,
'providerModelName', provider_model_name,
'modelType', model_type,
'modelAlias', provider_model_name,
'displayName', display_name,
'capabilities', capabilities,
'baseBillingConfig', base_billing_config,
'defaultRateLimitPolicy', default_rate_limit_policy,
'pricingRuleSetId', COALESCE(pricing_rule_set_id::text, ''),
'runtimePolicySetId', COALESCE(runtime_policy_set_id::text, ''),
'runtimePolicyOverride', runtime_policy_override,
'metadata', metadata,
'pricingVersion', pricing_version,
'status', 'active'
) AS default_snapshot
FROM payload
)
INSERT INTO base_model_catalog (
provider_id,
provider_key,
canonical_model_key,
invocation_name,
provider_model_name,
model_type,
display_name,
capabilities,
base_billing_config,
default_rate_limit_policy,
pricing_rule_set_id,
runtime_policy_set_id,
runtime_policy_override,
metadata,
catalog_type,
default_snapshot,
pricing_version,
status
)
SELECT
provider_id,
provider_key,
canonical_model_key,
provider_model_name,
provider_model_name,
model_type,
display_name,
capabilities,
base_billing_config,
default_rate_limit_policy,
pricing_rule_set_id,
runtime_policy_set_id,
runtime_policy_override,
metadata,
'system',
default_snapshot,
pricing_version,
'active'
FROM snapshot
ON CONFLICT (canonical_model_key) DO UPDATE
SET provider_id = EXCLUDED.provider_id,
provider_key = EXCLUDED.provider_key,
invocation_name = CASE
WHEN base_model_catalog.customized_at IS NULL THEN EXCLUDED.invocation_name
ELSE base_model_catalog.invocation_name
END,
provider_model_name = CASE
WHEN base_model_catalog.customized_at IS NULL THEN EXCLUDED.provider_model_name
ELSE base_model_catalog.provider_model_name
END,
model_type = CASE
WHEN base_model_catalog.customized_at IS NULL THEN EXCLUDED.model_type
ELSE base_model_catalog.model_type
END,
display_name = CASE
WHEN base_model_catalog.customized_at IS NULL THEN EXCLUDED.display_name
ELSE base_model_catalog.display_name
END,
capabilities = CASE
WHEN base_model_catalog.customized_at IS NULL THEN EXCLUDED.capabilities
ELSE base_model_catalog.capabilities
END,
base_billing_config = CASE
WHEN base_model_catalog.customized_at IS NULL THEN EXCLUDED.base_billing_config
ELSE base_model_catalog.base_billing_config
END,
default_rate_limit_policy = CASE
WHEN base_model_catalog.customized_at IS NULL THEN EXCLUDED.default_rate_limit_policy
ELSE base_model_catalog.default_rate_limit_policy
END,
pricing_rule_set_id = COALESCE(base_model_catalog.pricing_rule_set_id, EXCLUDED.pricing_rule_set_id),
runtime_policy_set_id = COALESCE(base_model_catalog.runtime_policy_set_id, EXCLUDED.runtime_policy_set_id),
runtime_policy_override = CASE
WHEN base_model_catalog.customized_at IS NULL THEN EXCLUDED.runtime_policy_override
ELSE base_model_catalog.runtime_policy_override
END,
metadata = CASE
WHEN base_model_catalog.customized_at IS NULL THEN EXCLUDED.metadata
ELSE base_model_catalog.metadata
END,
catalog_type = 'system',
default_snapshot = EXCLUDED.default_snapshot,
pricing_version = CASE
WHEN base_model_catalog.customized_at IS NULL THEN EXCLUDED.pricing_version
ELSE base_model_catalog.pricing_version
END,
status = CASE
WHEN base_model_catalog.customized_at IS NULL THEN 'active'
ELSE base_model_catalog.status
END,
updated_at = now();
INSERT INTO platform_models (
platform_id,
base_model_id,
model_name,
provider_model_name,
model_alias,
model_type,
display_name,
capabilities,
pricing_mode,
billing_config,
retry_policy,
rate_limit_policy,
enabled
)
SELECT
platform.id,
base_model.id,
base_model.invocation_name,
base_model.provider_model_name,
base_model.invocation_name,
base_model.model_type,
base_model.display_name,
base_model.capabilities,
'inherit_discount',
base_model.base_billing_config,
'{"enabled":true,"maxAttempts":1}'::jsonb,
base_model.default_rate_limit_policy,
true
FROM integration_platforms platform
JOIN base_model_catalog base_model
ON base_model.provider_key = platform.provider
WHERE base_model.canonical_model_key IN (
'easyai:Kimi-K3',
'volces-openai:kimi-k3',
'easyai:Qwen3.8-Max-Preview',
'aliyun-bailian-openai:qwen3.8-max-preview'
)
AND platform.deleted_at IS NULL
ON CONFLICT (platform_id, model_name) DO UPDATE
SET base_model_id = EXCLUDED.base_model_id,
provider_model_name = EXCLUDED.provider_model_name,
model_alias = EXCLUDED.model_alias,
display_name = EXCLUDED.display_name,
model_type = EXCLUDED.model_type,
capabilities = EXCLUDED.capabilities,
pricing_mode = EXCLUDED.pricing_mode,
billing_config = EXCLUDED.billing_config,
retry_policy = EXCLUDED.retry_policy,
rate_limit_policy = EXCLUDED.rate_limit_policy,
enabled = EXCLUDED.enabled,
updated_at = now();