mirror of
https://git.openapi.site/https://github.com/desirecore/config-center.git
synced 2026-10-21 07:04:11 +08:00
feat(compute): update current model catalog (#106)
* feat(compute): update current model catalog * feat(compute): add MiMo V2.6 model profiles
This commit is contained in:
GitHub
@@ -105,7 +105,7 @@ describe('真实数据全量校验', () => {
|
||||
if (spec.routing) routed.push(spec)
|
||||
}
|
||||
}
|
||||
assert.equal(routed.length, 48)
|
||||
assert.equal(routed.length, 58)
|
||||
assert.equal(routed.every((spec) => spec.routing.reasoning.supportedModes.includes(spec.routing.reasoning.defaultMode)), true)
|
||||
assert.equal(routed.every((spec) => Array.isArray(spec.spec.capabilities)), true)
|
||||
})
|
||||
@@ -151,11 +151,14 @@ describe('真实数据全量校验', () => {
|
||||
.map((model) => model.modelName)
|
||||
|
||||
assert.deepEqual(enabled(anthropic), [
|
||||
'claude-opus-5-5',
|
||||
'claude-fable-5-1',
|
||||
'claude-opus-5',
|
||||
'claude-sonnet-5',
|
||||
])
|
||||
assert.deepEqual(enabled(openai), [
|
||||
'gpt-6-sol',
|
||||
'gpt-6-luna',
|
||||
'gpt-6-astra',
|
||||
'gpt-5.5',
|
||||
'gpt-5.5-pro',
|
||||
@@ -515,32 +518,148 @@ describe('真实数据全量校验', () => {
|
||||
}
|
||||
})
|
||||
|
||||
it('Qwen3.8 Max Preview 应在 Token Plan 中提供完整的推理与视觉规格', () => {
|
||||
it('MiMo V2.6 三款官方 API 模型应保持全模态规格与接入差异', () => {
|
||||
const provider = JSON.parse(readFileSync(join(ROOT, 'compute', 'providers', 'xiaomi.json'), 'utf8'))
|
||||
const specFile = JSON.parse(readFileSync(join(ROOT, 'compute', 'model-specs', 'xiaomi.json'), 'utf8'))
|
||||
const cases = [
|
||||
['mimo-v2.6-pro', 3, 6, 0.025, 'flagship'],
|
||||
['mimo-v2.6-flash', 1, 2, 0.02, 'balanced'],
|
||||
['mimo-v2.6-pro-ultraspeed', 30, 60, 0.25, null],
|
||||
]
|
||||
|
||||
for (const [id, inputPrice, outputPrice, cachedInputPrice, tier] of cases) {
|
||||
const models = provider.models.filter((item) => item.modelName === id)
|
||||
const specs = specFile.specs.filter((item) => item.id === id)
|
||||
assert.equal(models.length, 1, `Provider 缺少或重复 ${id}`)
|
||||
assert.equal(specs.length, 1, `model-specs 缺少或重复 ${id}`)
|
||||
const model = models[0]
|
||||
const modelSpec = specs[0]
|
||||
assert.deepEqual(modelSpec.match.exact, [id])
|
||||
assert.equal(model.contextWindow, 1000000)
|
||||
assert.equal(model.maxOutputTokens, 131072)
|
||||
assert.equal(modelSpec.spec.contextWindow, model.contextWindow)
|
||||
assert.equal(modelSpec.spec.maxOutputTokens, model.maxOutputTokens)
|
||||
assert.equal(model.inputPrice, inputPrice)
|
||||
assert.equal(model.outputPrice, outputPrice)
|
||||
assert.equal(model.extra.cachedInputPrice, cachedInputPrice)
|
||||
for (const capability of ['vision', 'audio_understanding', 'video_understanding', 'tool_use']) {
|
||||
assert.ok(model.capabilities.includes(capability), `${id} 缺少 ${capability}`)
|
||||
assert.ok(modelSpec.spec.capabilities.includes(capability), `${id} 规格缺少 ${capability}`)
|
||||
}
|
||||
assert.equal(modelSpec.spec.extra.thinkingDefault, true)
|
||||
assert.equal(modelSpec.routing?.tier ?? null, tier)
|
||||
}
|
||||
|
||||
const previous = provider.models.find((item) => item.modelName === 'mimo-v2.5-pro')
|
||||
assert.ok(previous, '尚未到官方下线日，应保留 mimo-v2.5-pro')
|
||||
assert.equal(previous.maxOutputTokens, 131072)
|
||||
})
|
||||
|
||||
it('Token Plan 应使用 Qwen3.8 正式版 ID 并淘汰 Preview 预置', () => {
|
||||
const tokenPlan = JSON.parse(readFileSync(join(ROOT, 'compute', 'coding-plans', 'dashscope-token-plan.json'), 'utf8'))
|
||||
const apiProvider = JSON.parse(readFileSync(join(ROOT, 'compute', 'providers', 'dashscope.json'), 'utf8'))
|
||||
const specFile = JSON.parse(readFileSync(join(ROOT, 'compute', 'model-specs', 'qwen.json'), 'utf8'))
|
||||
const modelId = 'qwen3.8-max-preview'
|
||||
const model = tokenPlan.models.find((item) => item.modelName === modelId)
|
||||
assert.equal(tokenPlan.models.some((item) => item.modelName === 'qwen3.8-max-preview'), false)
|
||||
assert.equal(apiProvider.models.some((item) => item.modelName === 'qwen3.8-max-preview'), false)
|
||||
assert.ok(tokenPlan.tombstones.includes('qwen3.8-max-preview'))
|
||||
assert.ok(apiProvider.tombstones.includes('qwen3.8-max-preview'))
|
||||
for (const modelId of ['qwen3.8-max', 'qwen3.8-flash']) {
|
||||
const model = tokenPlan.models.find((item) => item.modelName === modelId)
|
||||
const modelSpec = specFile.specs.find((item) => item.id === modelId)
|
||||
assert.ok(model, `Token Plan 缺少 ${modelId}`)
|
||||
assert.ok(modelSpec, `model-specs 缺少 ${modelId}`)
|
||||
assert.equal(model.contextWindow, 1000000)
|
||||
assert.equal(model.maxOutputTokens, 131072)
|
||||
assert.ok(model.capabilities.includes('vision'))
|
||||
assert.deepEqual(model.extra.reasoning.supportedEfforts, ['low', 'medium', 'xhigh'])
|
||||
assert.equal(model.extra.reasoning.defaultEffort, 'xhigh')
|
||||
assert.equal(model.extra.thinkingOnly, undefined)
|
||||
assert.equal(model.extra.preserveThinkingDefault, true)
|
||||
assert.equal(modelSpec.spec.maxOutputTokens, 131072)
|
||||
const apiModel = apiProvider.models.find((item) => item.modelName === modelId)
|
||||
assert.deepEqual(apiModel.extra.reasoning.supportedEfforts, ['low', 'medium', 'xhigh'])
|
||||
assert.equal(apiModel.extra.preserveThinkingDefault, true)
|
||||
}
|
||||
})
|
||||
|
||||
assert.ok(model, `Token Plan 缺少 ${modelId}`)
|
||||
assert.equal(model.contextWindow, 983616)
|
||||
assert.equal(model.defaultTemperature, 0.6)
|
||||
assert.ok(model.capabilities.includes('reasoning'))
|
||||
assert.ok(model.capabilities.includes('vision'))
|
||||
assert.ok(model.serviceType.includes('reasoning'))
|
||||
assert.ok(model.serviceType.includes('vision'))
|
||||
assert.deepEqual(model.extra.reasoning.supportedEfforts, ['low', 'high', 'xhigh'])
|
||||
assert.equal(model.extra.reasoning.defaultEffort, 'xhigh')
|
||||
assert.equal(model.extra.thinkingOnly, true)
|
||||
assert.equal(model.extra.thinkingMaxTokens, 262144)
|
||||
assert.equal(model.extra.preserveThinkingDefault, true)
|
||||
assert.equal(model.extra.supportsParallelToolCalls, false)
|
||||
it('新增主流模型应在官方 Provider 与共享规格中保持 ID、窗口和推理配置一致', () => {
|
||||
const cases = [
|
||||
['openai', 'openai', 'gpt-6-sol', 1050000, 128000, 2, 10, 'medium'],
|
||||
['openai', 'openai', 'gpt-6-luna', 1050000, 128000, 0.1, 0.5, 'medium'],
|
||||
['anthropic', 'anthropic', 'claude-opus-5-5', 1000000, 128000, 4, 20, 'medium'],
|
||||
['google', 'google', 'gemini-3.8-flash', 1048576, 65536, 0.75, 3.75, null],
|
||||
['google', 'google', 'gemini-3.7-flash', 1048576, 65536, 0.75, 3.75, null],
|
||||
['google', 'google', 'gemini-3.5-flash-lite', 1048576, 65536, 0.3, 2.5, null],
|
||||
['dashscope', 'qwen', 'qwen3.8-max', 1000000, 131072, 12, 36, null],
|
||||
['dashscope', 'qwen', 'qwen3.8-flash', 1000000, 131072, 0.8, 2.7, null],
|
||||
['xai', 'xai', 'grok-4.7', 500000, undefined, 2, 6, 'high'],
|
||||
]
|
||||
|
||||
const specs = specFile.specs.filter((item) => item.id === modelId)
|
||||
assert.equal(specs.length, 1, `model-specs 中 ${modelId} 应且仅应有一条规格`)
|
||||
assert.equal(specs[0].spec.contextWindow, 1000000)
|
||||
assert.equal(specs[0].spec.defaultTemperature, 0.6)
|
||||
assert.equal(specs[0].spec.supportsReasoning, true)
|
||||
assert.ok(specs[0].spec.capabilities.includes('vision'))
|
||||
for (const [providerName, specName, id, contextWindow, maxOutputTokens, inputPrice, outputPrice, defaultEffort] of cases) {
|
||||
const provider = JSON.parse(readFileSync(join(ROOT, 'compute', 'providers', `${providerName}.json`), 'utf8'))
|
||||
const specFile = JSON.parse(readFileSync(join(ROOT, 'compute', 'model-specs', `${specName}.json`), 'utf8'))
|
||||
const models = provider.models.filter((item) => item.modelName === id)
|
||||
const specs = specFile.specs.filter((item) => item.id === id)
|
||||
assert.equal(models.length, 1, `${providerName} 缺少或重复 ${id}`)
|
||||
assert.equal(specs.length, 1, `${specName} 缺少或重复 ${id}`)
|
||||
assert.equal(models[0].contextWindow, contextWindow)
|
||||
assert.equal(specs[0].spec.contextWindow, contextWindow)
|
||||
assert.equal(models[0].maxOutputTokens, maxOutputTokens)
|
||||
assert.equal(specs[0].spec.maxOutputTokens, maxOutputTokens)
|
||||
assert.equal(models[0].inputPrice, inputPrice)
|
||||
assert.equal(models[0].outputPrice, outputPrice)
|
||||
assert.equal(specs[0].routing.reasoning.supportedModes.includes(specs[0].routing.reasoning.defaultMode), true)
|
||||
if (defaultEffort) assert.equal(models[0].extra.reasoning.defaultEffort, defaultEffort)
|
||||
}
|
||||
|
||||
const anthropic = JSON.parse(readFileSync(join(ROOT, 'compute', 'providers', 'anthropic.json'), 'utf8'))
|
||||
const opus = anthropic.models.find((item) => item.modelName === 'claude-opus-5-5')
|
||||
assert.equal(opus.extra.adaptiveThinking, true)
|
||||
assert.equal(opus.extra.thinkingOnly, true)
|
||||
assert.equal(opus.extra.forcedToolChoiceUnsupported, true)
|
||||
assert.equal(opus.extra.cachePricing.read, 0.2)
|
||||
|
||||
const xai = JSON.parse(readFileSync(join(ROOT, 'compute', 'providers', 'xai.json'), 'utf8'))
|
||||
assert.deepEqual(xai.models.find((item) => item.modelName === 'grok-4.7').extra.reasoning.supportedEfforts, ['low', 'medium', 'high', 'xhigh'])
|
||||
|
||||
const codex = JSON.parse(readFileSync(join(ROOT, 'compute', 'providers', 'openai-codex.json'), 'utf8'))
|
||||
for (const id of ['gpt-6-sol', 'gpt-6-luna']) {
|
||||
const model = codex.models.find((item) => item.modelName === id)
|
||||
assert.ok(model, `Codex 订阅 Provider 缺少 ${id}`)
|
||||
assert.equal(model.source, 'preset')
|
||||
assert.equal(model.extra.reasoning.defaultEffort, 'medium')
|
||||
}
|
||||
})
|
||||
|
||||
it('GLM-5.3-FlashX 仅加入共享规格，避免假定中国区智谱 API 已提供该 ID', () => {
|
||||
const provider = JSON.parse(readFileSync(join(ROOT, 'compute', 'providers', 'zhipu.json'), 'utf8'))
|
||||
const specs = JSON.parse(readFileSync(join(ROOT, 'compute', 'model-specs', 'zhipu.json'), 'utf8')).specs
|
||||
assert.equal(provider.models.some((item) => item.modelName === 'glm-5.3-flashx'), false)
|
||||
const spec = specs.find((item) => item.id === 'glm-5.3-flashx')
|
||||
assert.ok(spec)
|
||||
assert.deepEqual(spec.match.exact, ['glm-5.3-flashx', 'z-ai/glm-5.3-flashx'])
|
||||
assert.equal(spec.spec.contextWindow, 1000000)
|
||||
assert.equal(spec.spec.maxOutputTokens, 128000)
|
||||
assert.ok(spec.spec.capabilities.includes('vision'))
|
||||
})
|
||||
|
||||
it('GPT Image 2.5 双型号应使用图像生成接入，并与 GPT Image 2 精确区分', () => {
|
||||
const provider = JSON.parse(readFileSync(join(ROOT, 'compute', 'providers', 'openai.json'), 'utf8'))
|
||||
const specs = JSON.parse(readFileSync(join(ROOT, 'compute', 'model-specs', 'openai.json'), 'utf8')).specs
|
||||
for (const id of ['gpt-image-2.5-sunburst', 'gpt-image-2.5-flare']) {
|
||||
const model = provider.models.find((item) => item.modelName === id)
|
||||
const spec = specs.find((item) => item.id === id)
|
||||
assert.ok(model, `OpenAI Provider 缺少 ${id}`)
|
||||
assert.ok(spec, `model-specs 缺少 ${id}`)
|
||||
assert.deepEqual(model.serviceType, ['image_gen'])
|
||||
assert.deepEqual(spec.match.exact, [id])
|
||||
assert.equal(model.inputPrice, 5)
|
||||
assert.equal(model.extra.imageInputPrice, 8)
|
||||
assert.equal(model.outputPrice, 30)
|
||||
}
|
||||
const older = specs.find((item) => item.id === 'gpt-image-2')
|
||||
assert.deepEqual(older.match.exact, ['gpt-image-2'])
|
||||
assert.equal('patterns' in older.match, false)
|
||||
})
|
||||
|
||||
it('Qwen3.8 Flash 应提供完整的多模态推理规格', () => {
|
||||
@@ -618,7 +737,8 @@ describe('真实数据全量校验', () => {
|
||||
assert.equal('patterns' in deepseekFlash.match, false)
|
||||
assert.equal('patterns' in deepseekFlash0731.match, false)
|
||||
assert.equal(qwenBase.spec.contextWindow, qwenPreview.spec.contextWindow)
|
||||
assert.equal(qwenBase.spec.maxOutputTokens, qwenPreview.spec.maxOutputTokens)
|
||||
assert.equal(qwenBase.spec.maxOutputTokens, 131072)
|
||||
assert.equal(qwenPreview.spec.maxOutputTokens, 65536)
|
||||
assert.equal(deepseekBase.spec.contextWindow, deepseek0813.spec.contextWindow)
|
||||
assert.equal(deepseekBase.spec.maxOutputTokens, deepseek0813.spec.maxOutputTokens)
|
||||
assert.equal(deepseekFlash.spec.contextWindow, deepseekFlash0731.spec.contextWindow)
|
||||
@@ -632,16 +752,16 @@ describe('真实数据全量校验', () => {
|
||||
// 两处漏写都会回退成 budget_tokens（desirecore#2940）。Opus 4.6 / Sonnet 4.6 仍接受
|
||||
// budget_tokens（仅弃用），不在强制范围内。新代际若同样拒绝 budget_tokens，先扩展这里的正则。
|
||||
const adaptiveGeneration = /^claude-(?:fable|mythos)(?:$|-)|^claude-(?:opus|sonnet)-(?:[5-9]|[1-9]\d)(?:$|-)|^claude-opus-4-[78](?:$|-)/
|
||||
const thinkingOnlyGeneration = /^claude-(?:fable|mythos)(?:$|-)/
|
||||
const thinkingOnlyGeneration = /^claude-(?:fable|mythos)(?:$|-)|^claude-opus-5-5(?:$|-)/
|
||||
// 与客户端匹配器同口径：小写、去 vendor 前缀、统一分隔符
|
||||
const normalizeId = (id) => id.toLowerCase().trim().split('/').at(-1).replace(/[._:\s]+/g, '-')
|
||||
const specsDir = join(ROOT, 'compute', 'model-specs')
|
||||
const allSpecs = JSON.parse(readFileSync(join(specsDir, '_index.json'), 'utf8')).order
|
||||
.flatMap((name) => JSON.parse(readFileSync(join(specsDir, `${name}.json`), 'utf8')).specs)
|
||||
|
||||
it('规格按代际声明 adaptiveThinking，只有 Fable/Mythos 声明思考恒开', () => {
|
||||
it('规格按代际声明 adaptiveThinking，并标记始终开启思考的模型', () => {
|
||||
const adaptiveSpecIds = allSpecs.map((item) => item.id).filter((id) => adaptiveGeneration.test(normalizeId(id)))
|
||||
for (const id of ['claude-fable-5-1', 'claude-opus-5', 'claude-opus-4-8', 'claude-opus-4-7', 'claude-sonnet-5']) {
|
||||
for (const id of ['claude-fable-5-1', 'claude-opus-5-5', 'claude-opus-5', 'claude-opus-4-8', 'claude-opus-4-7', 'claude-sonnet-5']) {
|
||||
assert.ok(adaptiveSpecIds.includes(id), `model-specs 缺少 ${id}`)
|
||||
}
|
||||
for (const modelSpec of allSpecs) {
|
||||
@@ -715,10 +835,10 @@ describe('真实数据全量校验', () => {
|
||||
// 只做单向断言（已知拒绝的代际必须声明），新代际按官方文档补声明即可，不维护反向代际表。
|
||||
// 订阅（claude-oauth）请求由 Claude CLI 自建，客户端 compat 层对这两类参数显式 fail closed，不依赖声明。
|
||||
const samplingRemoved = /^claude-(?:fable|mythos)-|^claude-(?:opus|sonnet)-5(?:$|-)|^claude-opus-4-[78](?:$|-)/
|
||||
const forcedToolChoiceRemoved = /^claude-(?:fable|mythos)-5-1(?:$|-)/
|
||||
const forcedToolChoiceRemoved = /^claude-(?:fable|mythos)-5-1(?:$|-)|^claude-opus-5-5(?:$|-)/
|
||||
|
||||
const specFile = JSON.parse(readFileSync(join(ROOT, 'compute', 'model-specs', 'anthropic.json'), 'utf8'))
|
||||
for (const id of ['claude-fable-5-1', 'claude-opus-5', 'claude-opus-4-8', 'claude-opus-4-7', 'claude-sonnet-5']) {
|
||||
for (const id of ['claude-fable-5-1', 'claude-opus-5-5', 'claude-opus-5', 'claude-opus-4-8', 'claude-opus-4-7', 'claude-sonnet-5']) {
|
||||
assert.ok(specFile.specs.some((item) => item.id === id), `model-specs 缺少 ${id}`)
|
||||
}
|
||||
for (const modelSpec of specFile.specs) {
|
||||
|
||||
Reference in New Issue
Block a user