fix(models): correct fast pricing and online sync

frontend/src/api/__tests__/models-dev-pricing.spec.ts

@@ -63,6 +63,14 @@ describe('buildModelsDevTieredPricing', () => {
     })
   })
 
+  it('fails closed when a legacy long-context copy has no authoritative tier boundary', () => {
+    expect(buildModelsDevTieredPricing({
+      input: 5,
+      output: 30,
+      context_over_200k: { input: 10, output: 45 },
+    })).toBeNull()
+  })
+
   it('allows special token dimensions only when they use the base token price', () => {
     expect(buildModelsDevTieredPricing({
       input: 1,
@@ -214,7 +222,7 @@ describe('resolveModelsDevTieredPricing', () => {
     expect(resolveModelsDevTieredPricing('openai', 'gpt-5.6-sol', undefined)).toBeNull()
   })
 
-  it('keeps a models.dev fast cost as an explicit Priority catalog when bands differ', () => {
+  it('collapses a flat models.dev Fast cost to the Standard first band multiplier', () => {
     expect(resolveModelsDevTieredPricing('openai', 'gpt-5.6-sol', {
       input: 5,
       output: 30,
@@ -235,12 +243,81 @@ describe('resolveModelsDevTieredPricing', () => {
       ],
       processing_tiers: {
         priority: {
-          tiers: [{ up_to: null, input_price_per_1m: 10, output_price_per_1m: 60 }],
+          price_multiplier: 2,
         },
       },
     })
   })
 
+  it('maps the GPT-5.6 Fast cost and ignores the legacy context copy', () => {
+    expect(resolveModelsDevTieredPricing('openai', 'gpt-5.6-sol', {
+      input: 5,
+      output: 30,
+      cache_read: 0.5,
+      cache_write: 6.25,
+      tiers: [{
+        input: 10,
+        output: 45,
+        cache_read: 1,
+        cache_write: 12.5,
+        tier: { type: 'context', size: 272_000 },
+      }],
+      context_over_200k: {
+        input: 10,
+        output: 45,
+        cache_read: 1,
+        cache_write: 12.5,
+      },
+    }, {
+      fast: {
+        cost: {
+          input: 10,
+          output: 60,
+          cache_read: 1,
+          cache_write: 12.5,
+        },
+        provider: { body: { service_tier: 'priority' } },
+      },
+    })).toEqual({
+      tiers: [
+        {
+          up_to: 271_999,
+          input_price_per_1m: 5,
+          output_price_per_1m: 30,
+          cache_creation_price_per_1m: 6.25,
+          cache_read_price_per_1m: 0.5,
+        },
+        {
+          up_to: null,
+          input_price_per_1m: 10,
+          output_price_per_1m: 45,
+          cache_creation_price_per_1m: 12.5,
+          cache_read_price_per_1m: 1,
+        },
+      ],
+      processing_tiers: {
+        priority: { price_multiplier: 2 },
+      },
+    })
+  })
+
+  it('fails closed when an experimental mode tries to supply context tiers', () => {
+    expect(resolveModelsDevTieredPricing('openai', 'gpt-5.6-sol', {
+      input: 5,
+      output: 30,
+      tiers: [{ input: 10, output: 45, tier: { type: 'context', size: 272_000 } }],
+    }, {
+      fast: {
+        cost: {
+          input: 10,
+          output: 60,
+          tiers: [{ input: 20, output: 90, tier: { type: 'context', size: 272_000 } }],
+        },
+        provider: { body: { service_tier: 'priority' } },
+      },
+    })?.processing_tiers).toBeUndefined()
+  })
+
   it('uses a multiplier only when every fast price has the same ratio', () => {
     expect(resolveModelsDevTieredPricing('anthropic', 'claude-opus-4.8', {
       input: 5,
@@ -269,6 +346,27 @@ describe('resolveModelsDevTieredPricing', () => {
     })
   })
 
+  it('keeps a non-uniform Fast cost as one explicit unbounded band', () => {
+    expect(resolveModelsDevTieredPricing('openai', 'future-model', {
+      input: 5,
+      output: 30,
+      tiers: [{
+        input: 10,
+        output: 45,
+        tier: { type: 'context', size: 272_000 },
+      }],
+    }, {
+      fast: {
+        cost: { input: 10, output: 75 },
+        provider: { body: { service_tier: 'priority' } },
+      },
+    })?.processing_tiers).toEqual({
+      priority: {
+        tiers: [{ up_to: null, input_price_per_1m: 10, output_price_per_1m: 75 }],
+      },
+    })
+  })
+
   it('uses the standard catalog when an imported tier has a zero default ratio', () => {
     expect(resolveModelsDevTieredPricing('openai', 'gpt-5.6-sol', {
       input: 5,

frontend/src/api/__tests__/models-dev.spec.ts

@@ -2,18 +2,21 @@ import { beforeEach, describe, expect, it, vi } from 'vitest'
 
 const apiMocks = vi.hoisted(() => ({
   get: vi.fn(),
+  delete: vi.fn(),
 }))
 
 vi.mock('@/api/client', () => ({
-  default: { get: apiMocks.get },
+  default: { get: apiMocks.get, delete: apiMocks.delete },
 }))
 
-import { clearModelsDevCache, getModelsDevList } from '@/api/models-dev'
+import { clearModelsDevCache, getModelsDevList, refreshModelsDevList } from '@/api/models-dev'
 
 beforeEach(() => {
   clearModelsDevCache()
   localStorage.clear()
   apiMocks.get.mockReset()
+  apiMocks.delete.mockReset()
+  apiMocks.delete.mockResolvedValue({ data: { cleared: true } })
 })
 
 describe('getModelsDevList', () => {
@@ -34,7 +37,15 @@ describe('getModelsDevList', () => {
                 input: ['text', 'image'],
                 output: ['text', 'image'],
               },
-              cost: { input: 2, output: 4 },
+              cost: {
+                input: 2,
+                output: 4,
+                tiers: [{
+                  input: 4,
+                  output: 8,
+                  tier: { type: 'context', size: 100_000 },
+                }],
+              },
               experimental: {
                 modes: {
                   fast: {
@@ -86,4 +97,29 @@ describe('getModelsDevList', () => {
     })
     expect(audioPriced?.tieredPricing).toBeUndefined()
   })
+
+  it('clears the gateway cache before rebuilding the online model list', async () => {
+    apiMocks.get.mockResolvedValue({
+      data: {
+        openai: {
+          name: 'OpenAI',
+          official: true,
+          models: {
+            'gpt-test': {
+              id: 'gpt-test',
+              name: 'GPT Test',
+              cost: { input: 2, output: 4 },
+            },
+          },
+        },
+      },
+    })
+
+    await getModelsDevList(false)
+    await refreshModelsDevList(false)
+
+    expect(apiMocks.delete).toHaveBeenCalledOnce()
+    expect(apiMocks.delete).toHaveBeenCalledWith('/api/admin/models/external/cache')
+    expect(apiMocks.get).toHaveBeenCalledTimes(2)
+  })
 })