feat(kimi): add Kimi K3 context variants

docs/integrations/overview.md

@@ -126,6 +126,21 @@ fields. Only add `reasoning` metadata when the exact route/model request shape,
 accepted levels, and disable behavior have been verified. See
 `docs/integrations/reasoning-effort.md`.
 
+### Moonshot and Kimi K3 catalogs
+
+The direct Moonshot API exposes Kimi K3 as `kimi-k3`, with a 1,048,576-token
+context window, 32,768 maximum output tokens, and `reasoning_effort` levels
+`low`, `high`, and `max` (defaulting to `max`). Kimi Code K3 uses the same
+controls for two catalog selections: `k3` provides the 1M Allegretto+ window, while
+`k3-256k` keeps Moderato+ sessions within their 256K limit.
+
+Kimi Code also lists `kimi-for-coding-highspeed` for eligible Allegretto+
+subscriptions. [Kimi documents](https://www.kimi.com/code/docs/en/kimi-code/models.html)
+it as HighSpeed with approximately 6× output speed and 3× quota usage; plan
+availability can vary. Retain the selected catalog ID in client-side routing so
+its route-specific limits and capabilities are not lost when the outbound API
+model is normalized.
+
 ## Descriptor Authoring Pattern
 
 Normal descriptor files should:

src/integrations/gateways/kimi-code.ts

@@ -5,6 +5,8 @@ export default defineGateway({
   label: 'Moonshot AI - Kimi Code',
   category: 'hosted',
   defaultBaseUrl: 'https://api.kimi.com/coding/v1',
+  // K3 requires a Moderato subscription. Keep the all-members K2.7 model as
+  // the default; eligible users can explicitly select either K3 variant.
   defaultModel: 'kimi-for-coding',
   supportsModelRouting: true,
   setup: {
@@ -33,8 +35,11 @@ export default defineGateway({
   catalog: {
     source: 'static',
     models: [
+      { id: 'k3', apiName: 'k3', label: 'Kimi K3 (1M)', modelDescriptorId: 'k3', contextWindow: 1_048_576, maxOutputTokens: 32_768, capabilities: { supportsVision: true, supportsFunctionCalling: true, supportsJsonMode: true, supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'high', 'max'], defaultLevel: 'max', wireFormat: 'reasoning_effort', disableFormat: 'thinking_type_disabled' }, notes: 'Allegretto+' },
+      { id: 'k3-256k', apiName: 'k3', label: 'Kimi K3 (256K)', modelDescriptorId: 'k3', contextWindow: 262_144, maxOutputTokens: 32_768, capabilities: { supportsVision: true, supportsFunctionCalling: true, supportsJsonMode: true, supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'high', 'max'], defaultLevel: 'max', wireFormat: 'reasoning_effort', disableFormat: 'thinking_type_disabled' }, notes: 'Moderato+' },
       { id: 'kimi-k2.7-code', apiName: 'kimi-k2.7-code', aliases: ['moonshotai/kimi-k2.7-code'], label: 'Kimi K2.7 Code', modelDescriptorId: 'kimi-k2.7-code', contextWindow: 262_144, maxOutputTokens: 32_768, capabilities: { supportsVision: true, supportsFunctionCalling: true, supportsJsonMode: true, supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high'], defaultLevel: 'medium', wireFormat: 'reasoning_effort' } },
       { id: 'kimi-for-coding', apiName: 'kimi-for-coding', label: 'Kimi for Coding', modelDescriptorId: 'kimi-for-coding', contextWindow: 262_144, maxOutputTokens: 32_768, capabilities: { supportsVision: true, supportsFunctionCalling: true, supportsJsonMode: true, supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high'], defaultLevel: 'medium', wireFormat: 'reasoning_effort' } },
+      { id: 'kimi-for-coding-highspeed', apiName: 'kimi-for-coding-highspeed', label: 'Kimi K2.7 Code HighSpeed (Allegretto+)', modelDescriptorId: 'kimi-for-coding', contextWindow: 262_144, maxOutputTokens: 32_768, capabilities: { supportsVision: true, supportsFunctionCalling: true, supportsJsonMode: true, supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high'], defaultLevel: 'medium', wireFormat: 'reasoning_effort' }, notes: 'Requires Allegretto or above; 6x speed and 3x quota usage.' },
     ],
   },
   usage: { supported: false },

src/integrations/index.test.ts

@@ -99,6 +99,7 @@ describe('loaded registry validation', () => {
         const defaultModel = gateway.defaultModel?.trim()
         return !(gateway.catalog?.models ?? []).some(
           entry =>
+            entry.id === defaultModel ||
             entry.apiName === defaultModel ||
             entry.modelDescriptorId === defaultModel,
         )

src/integrations/models/kimi.test.ts

@@ -0,0 +1,8 @@
+import { expect, test } from 'bun:test'
+
+import { getModel, getModelsForBrand } from '../index.js'
+
+test('K3 belongs to the Kimi brand', () => {
+  expect(getModel('k3')?.brandId).toBe('kimi')
+  expect(getModelsForBrand('kimi').map(model => model.id)).toContain('k3')
+})

src/integrations/models/kimi.ts

@@ -29,6 +29,7 @@ function kimiModel(
 }
 
 export default [
+  kimiModel('k3', 'Kimi K3', 1_048_576, 32_768),
   kimiModel('kimi-for-coding', 'Kimi for Coding', 262_144, 32_768),
   kimiModel('kimi-k2.7-code', 'Kimi K2.7 Code', 262_144, 262_144),
   kimiModel('kimi-k2.6', 'Kimi K2.6', 262_144, 32_768),

src/integrations/runtimeMetadata.test.ts

@@ -304,6 +304,14 @@ describe('resolveOpenAIShimRuntimeContext - NVIDIA NIM GLM-5.2 (regression #1950
 
 describe('resolveOpenAIShimRuntimeContext - Moonshot and Kimi Code catalog metadata', () => {
   it('uses Moonshot direct catalog order, limits, and reasoning controls', () => {
+    expect(
+      resolveModelRuntimeLimits({
+        model: 'kimi-k3',
+        baseUrl: 'https://api.moonshot.ai/v1',
+        processEnv: { CLAUDE_CODE_USE_OPENAI: '1' },
+      }),
+    ).toEqual({ contextWindow: 1_048_576, maxOutputTokens: 32_768 })
+
     expect(
       resolveModelRuntimeLimits({
         model: 'kimi-k2.7-code',
@@ -336,6 +344,7 @@ describe('resolveOpenAIShimRuntimeContext - Moonshot and Kimi Code catalog metad
 
     expect(result.routeId).toBe('moonshot')
     expect(result.descriptor?.catalog?.models?.map(model => model.id)).toEqual([
+      'k3',
       'kimi-k2.7-code',
       'kimi-k2.6',
       'kimi-k2.5',
@@ -374,9 +383,40 @@ describe('resolveOpenAIShimRuntimeContext - Moonshot and Kimi Code catalog metad
 
     expect(result.routeId).toBe('kimi-code')
     expect(result.descriptor?.catalog?.models?.map(model => model.id)).toEqual([
+      'k3',
+      'k3-256k',
       'kimi-k2.7-code',
       'kimi-for-coding',
+      'kimi-for-coding-highspeed',
     ])
+    const k3 = resolveOpenAIShimRuntimeContext({
+      model: 'k3',
+      baseUrl: 'https://api.kimi.com/coding/v1',
+      processEnv: { CLAUDE_CODE_USE_OPENAI: '1' },
+    })
+    expect(k3.catalogEntry).toMatchObject({
+      id: 'k3',
+      contextWindow: 1_048_576,
+      label: 'Kimi K3 (1M)',
+    })
+    expect(k3.catalogEntry?.reasoning?.levels).toEqual(['low', 'high', 'max'])
+    expect(resolveModelRuntimeLimits({
+      model: 'k3-256k',
+      baseUrl: 'https://api.kimi.com/coding/v1',
+      processEnv: { CLAUDE_CODE_USE_OPENAI: '1' },
+    })).toEqual({ contextWindow: 262_144, maxOutputTokens: 32_768 })
+    const highspeed = resolveOpenAIShimRuntimeContext({
+      model: 'kimi-for-coding-highspeed',
+      baseUrl: 'https://api.kimi.com/coding/v1',
+      processEnv: { CLAUDE_CODE_USE_OPENAI: '1' },
+    })
+    expect(highspeed.catalogEntry).toMatchObject({
+      id: 'kimi-for-coding-highspeed',
+      apiName: 'kimi-for-coding-highspeed',
+      contextWindow: 262_144,
+      maxOutputTokens: 32_768,
+    })
+    expect(highspeed.catalogEntry?.reasoning?.levels).toEqual(['low', 'medium', 'high'])
     expect(result.catalogEntry?.id).toBe('kimi-for-coding')
     expect(result.catalogEntry?.reasoning?.levels).toEqual(['low', 'medium', 'high'])
     expect(result.catalogEntry?.reasoning?.defaultLevel).toBe('medium')

src/integrations/runtimeMetadata.ts

@@ -66,7 +66,10 @@ function matchesCatalogEntryModel(
   entry: ModelCatalogEntry,
   modelApiName: string,
 ): boolean {
-  if (entry.apiName.trim().toLowerCase() === modelApiName) {
+  if (
+    entry.apiName.trim().toLowerCase() === modelApiName ||
+    entry.id.trim().toLowerCase() === modelApiName
+  ) {
     return true
   }
 

src/integrations/vendors/moonshot.ts

@@ -34,6 +34,7 @@ export default defineVendor({
   catalog: {
     source: 'static',
     models: [
+      { id: 'k3', apiName: 'kimi-k3', label: 'Kimi K3', modelDescriptorId: 'k3', contextWindow: 1_048_576, maxOutputTokens: 32_768, capabilities: { supportsVision: true, supportsFunctionCalling: true, supportsJsonMode: true, supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'high', 'max'], defaultLevel: 'max', wireFormat: 'reasoning_effort' } },
       { id: 'kimi-k2.7-code', apiName: 'kimi-k2.7-code', aliases: ['moonshotai/kimi-k2.7-code'], label: 'Kimi K2.7 Code', modelDescriptorId: 'kimi-k2.7-code', contextWindow: 262_144, maxOutputTokens: 32_768, capabilities: { supportsVision: true, supportsFunctionCalling: true, supportsJsonMode: true, supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high'], defaultLevel: 'medium', wireFormat: 'reasoning_effort' } },
       { id: 'kimi-k2.6', apiName: 'kimi-k2.6', aliases: ['moonshotai/kimi-k2.6'], label: 'Kimi K2.6', modelDescriptorId: 'kimi-k2.6', contextWindow: 262_144, maxOutputTokens: 262_144, capabilities: { supportsVision: true, supportsFunctionCalling: true, supportsJsonMode: true, supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high'], defaultLevel: 'medium', wireFormat: 'reasoning_effort' } },
       { id: 'kimi-k2.5', apiName: 'kimi-k2.5', aliases: ['moonshotai/kimi-k2.5'], label: 'Kimi K2.5', modelDescriptorId: 'kimi-k2.5', contextWindow: 262_144, maxOutputTokens: 262_144, capabilities: { supportsVision: true, supportsFunctionCalling: true, supportsJsonMode: true, supportsReasoning: true }, reasoning: { mode: 'levels', levels: ['low', 'medium', 'high'], defaultLevel: 'medium', wireFormat: 'reasoning_effort' } },