From e4f94a4ee48b3cd2cd4349c2b3cf23a6b68af742 Mon Sep 17 00:00:00 2001
From: Siarhei Yarkavy <siarhei.yarkavy@gmail.com>
Date: Sat, 26 Sep 2026 13:04:09 +0300
Subject: [PATCH 1/8] fix: report 1050000-token window for GPT-6 Luna and Sol
 models

codemie-opencode overrode models.dev limits with heuristic defaults that had no gpt-6 branch, showing ~128k instead of the vendor-published 1050000.

Refs #581
---
 .../__tests__/opencode-gpt55-routing.test.ts  | 63 +++++++++++++++++
 .../opencode/opencode-dynamic-models.ts       |  6 +-
 .../opencode/opencode-model-configs.ts        | 68 +++++++++++++++++++
 .../plugins/pi/__tests__/pi.models.test.ts    | 32 +++++++++
 src/agents/plugins/pi/pi.models.ts            |  4 ++
 .../__tests__/vscode-models.test.ts           | 13 +++-
 .../proxy/connectors/__tests__/vscode.test.ts | 26 ++++---
 .../proxy/connectors/vscode-models.ts         | 22 ++++++
 8 files changed, 221 insertions(+), 13 deletions(-)

diff --git a/src/agents/plugins/__tests__/opencode-gpt55-routing.test.ts b/src/agents/plugins/__tests__/opencode-gpt55-routing.test.ts
index e9a5c0ea2..b415963df 100644
--- a/src/agents/plugins/__tests__/opencode-gpt55-routing.test.ts
+++ b/src/agents/plugins/__tests__/opencode-gpt55-routing.test.ts
@@ -165,3 +165,66 @@ describe('GPT-5.6 → Responses API routing', () => {
     expect(OPENCODE_MODEL_CONFIGS['gpt-5.6-sol-2026-07-09']!.limit.context).toBe(1050000);
   });
 });
+
+describe('GPT-6 → Responses API routing', () => {
+  let convertApiModelToOpenCodeConfig: typeof import('../opencode/opencode-dynamic-models.js').convertApiModelToOpenCodeConfig;
+  let OPENCODE_MODEL_CONFIGS: typeof import('../opencode/opencode-model-configs.js').OPENCODE_MODEL_CONFIGS;
+
+  beforeEach(async () => {
+    vi.resetModules();
+    ({ convertApiModelToOpenCodeConfig } = await import('../opencode/opencode-dynamic-models.js'));
+    ({ OPENCODE_MODEL_CONFIGS } = await import('../opencode/opencode-model-configs.js'));
+  });
+
+  // ── Dynamic path (live catalogue) ──────────────────────────────────────────
+
+  it('routes gpt-6-luna to Responses API via dynamic model conversion', () => {
+    const config = convertApiModelToOpenCodeConfig(makeLlmModel('gpt-6-luna'));
+    expect(config.use_responses_api).toBe(true);
+  });
+
+  it('routes openai.gpt-6-luna (vendor-prefixed) to Responses API via dynamic conversion', () => {
+    const config = convertApiModelToOpenCodeConfig(makeLlmModel('openai.gpt-6-luna'));
+    expect(config.use_responses_api).toBe(true);
+  });
+
+  it('routes gpt-6-sol-2026-09-22 (dated) to Responses API via dynamic conversion', () => {
+    const config = convertApiModelToOpenCodeConfig(makeLlmModel('gpt-6-sol-2026-09-22'));
+    expect(config.use_responses_api).toBe(true);
+  });
+
+  it('detects the gpt-6 family for vendor-prefixed ids', () => {
+    const config = convertApiModelToOpenCodeConfig(makeLlmModel('openai.gpt-6-sol'));
+    expect(config.family).toBe('gpt-6');
+  });
+
+  it('dynamic gpt-6-luna reports context limit of 1050000', () => {
+    const config = convertApiModelToOpenCodeConfig(makeLlmModel('gpt-6-luna'));
+    expect(config.limit.context).toBe(1050000);
+    expect(config.limit.output).toBe(128000);
+  });
+
+  it('dynamic openai.gpt-6-luna reports context limit of 1050000', () => {
+    const config = convertApiModelToOpenCodeConfig(makeLlmModel('openai.gpt-6-luna'));
+    expect(config.limit.context).toBe(1050000);
+  });
+
+  // ── Static fallback path (OPENCODE_MODEL_CONFIGS) ──────────────────────────
+
+  it('static config has gpt-6-luna with use_responses_api: true', () => {
+    expect(OPENCODE_MODEL_CONFIGS['gpt-6-luna']).toBeDefined();
+    expect(OPENCODE_MODEL_CONFIGS['gpt-6-luna']!.use_responses_api).toBe(true);
+  });
+
+  it('static config gpt-6-sol supports tool_call', () => {
+    expect(OPENCODE_MODEL_CONFIGS['gpt-6-sol']!.tool_call).toBe(true);
+  });
+
+  it('static config gpt-6-luna reports context limit of 1050000', () => {
+    expect(OPENCODE_MODEL_CONFIGS['gpt-6-luna']!.limit.context).toBe(1050000);
+  });
+
+  it('static config gpt-6-sol reports context limit of 1050000', () => {
+    expect(OPENCODE_MODEL_CONFIGS['gpt-6-sol']!.limit.context).toBe(1050000);
+  });
+});
diff --git a/src/agents/plugins/opencode/opencode-dynamic-models.ts b/src/agents/plugins/opencode/opencode-dynamic-models.ts
index 3ff431be7..209065422 100644
--- a/src/agents/plugins/opencode/opencode-dynamic-models.ts
+++ b/src/agents/plugins/opencode/opencode-dynamic-models.ts
@@ -28,7 +28,8 @@ import { logger } from '../../../utils/logger.js';
 //
 // Naming conventions observed in CodeMie deployments:
 //   Responses API  →  gpt-5-2-*, gpt-5.2-*, gpt-5.x-codex-*, gpt-5-x-codex-*,
-//                     gpt-5.4-*, gpt-5-4-*, gpt-5.5-*, gpt-5-5-*, gpt-5.6-*, gpt-5-6-*
+//                     gpt-5.4-*, gpt-5-4-*, gpt-5.5-*, gpt-5-5-*, gpt-5.6-*, gpt-5-6-*,
+//                     gpt-6-*, gpt-6.* (e.g. gpt-6-luna, openai.gpt-6-luna)
 //   Chat Completions → gpt-4*, gpt-5-<year>-*, o1/o3/o4*, gemini-*, claude-*, …
 //
 // Update this list whenever new Responses-API-only models are deployed.
@@ -45,6 +46,7 @@ const RESPONSES_API_MODEL_PATTERNS: RegExp[] = [
   /^gpt-5\.5-/,       // gpt-5.5-2026-04-24 — same Azure restriction as gpt-5.4
   /^gpt-5-5-/,        // hyphenated variant of gpt-5.5-*
   /gpt-5[.-]6/,       // gpt-5.6-* (e.g. openai.gpt-5.6-luna) — same Azure restriction (tools + reasoning_effort)
+  /gpt-6[.-]/,        // gpt-6-* (e.g. gpt-6-luna, openai.gpt-6-luna) — Responses API for built-in tools
 ];
 
 function isResponsesApiModel(id: string): boolean {
@@ -58,6 +60,7 @@ function detectFamily(id: string): string {
   if (id.startsWith('gemini')) return 'gemini-2';
   if (id.startsWith('gpt-4')) return 'gpt-4';
   if (id.startsWith('gpt-5') || /gpt-5[.-]6/.test(id)) return 'gpt-5';
+  if (id.startsWith('gpt-6') || /gpt-6[.-]/.test(id)) return 'gpt-6';
   if (/^o[134]-/.test(id) || id === 'o1') return 'openai-reasoning';
   if (id.startsWith('qwen')) return 'qwen3';
   if (id.startsWith('deepseek')) return 'deepseek';
@@ -77,6 +80,7 @@ function detectLimits(id: string, family: string): { context: number; output: nu
   if (id.startsWith('gpt-4o')) return { context: 128000, output: 16384 };
   if (id.startsWith('gpt-5.5') || id.startsWith('gpt-5-5')) return { context: 1050000, output: 128000 }; // Azure-published window for gpt-5.5
   if (/gpt-5[.-]6/.test(id)) return { context: 1050000, output: 128000 }; // Azure-published window for gpt-5.6
+  if (id.startsWith('gpt-6') || /gpt-6[.-]/.test(id)) return { context: 1050000, output: 128000 }; // Vendor-published window for gpt-6 (1,050,000 context, 128k output)
   if (id.startsWith('gpt-5')) return { context: 400000, output: 128000 };
   if (/^o[134]-/.test(id) || id === 'o1') return { context: 200000, output: 100000 };
   if (id.startsWith('qwen') || id.startsWith('moonshotai') || id.startsWith('kimi')) return { context: 262144, output: 131072 };
diff --git a/src/agents/plugins/opencode/opencode-model-configs.ts b/src/agents/plugins/opencode/opencode-model-configs.ts
index 81ff5e0d8..e87ac3418 100644
--- a/src/agents/plugins/opencode/opencode-model-configs.ts
+++ b/src/agents/plugins/opencode/opencode-model-configs.ts
@@ -288,6 +288,66 @@ export const OPENCODE_MODEL_CONFIGS: Record<string, OpenCodeModelConfig> = {
     }
   },
 
+  // ── GPT-6 Models (vendor-published: 1,050,000 context, 128k output, Responses API) ──
+  'gpt-6-luna': {
+    id: 'gpt-6-luna',
+    name: 'GPT-6 Luna',
+    displayName: 'GPT-6 Luna',
+    family: 'gpt-6',
+    tool_call: true,
+    reasoning: true,
+    attachment: true,
+    temperature: false,
+    structured_output: true,
+    use_responses_api: true,
+    modalities: {
+      input: ['text', 'image'],
+      output: ['text']
+    },
+    knowledge: '2026-05-18',
+    release_date: '2026-09-22',
+    last_updated: '2026-09-22',
+    open_weights: false,
+    cost: {
+      input: 0.10,
+      output: 0.50,
+      cache_read: 0.01
+    },
+    limit: {
+      context: 1050000,
+      output: 128000
+    }
+  },
+  'gpt-6-sol': {
+    id: 'gpt-6-sol',
+    name: 'GPT-6 Sol',
+    displayName: 'GPT-6 Sol',
+    family: 'gpt-6',
+    tool_call: true,
+    reasoning: true,
+    attachment: true,
+    temperature: false,
+    structured_output: true,
+    use_responses_api: true,
+    modalities: {
+      input: ['text', 'image'],
+      output: ['text']
+    },
+    knowledge: '2026-05-18',
+    release_date: '2026-09-22',
+    last_updated: '2026-09-22',
+    open_weights: false,
+    cost: {
+      input: 2,
+      output: 10,
+      cache_read: 0.20
+    },
+    limit: {
+      context: 1050000,
+      output: 128000
+    }
+  },
+
   // ── Claude Models ──────────────────────────────────────────────────
   'claude-4-5-sonnet': {
     id: 'claude-4-5-sonnet',
@@ -647,6 +707,14 @@ const MODEL_FAMILY_DEFAULTS: Record<string, Partial<OpenCodeModelConfig>> = {
     modalities: { input: ['text', 'image', 'audio', 'video'], output: ['text'] },
     limit: { context: 1048576, output: 65536 }
   },
+  'gpt-6': {
+    family: 'gpt-6',
+    reasoning: true,
+    attachment: true,
+    temperature: false,
+    modalities: { input: ['text', 'image'], output: ['text'] },
+    limit: { context: 1050000, output: 128000 }
+  },
   'gpt': {
     family: 'gpt-5',
     reasoning: true,
diff --git a/src/agents/plugins/pi/__tests__/pi.models.test.ts b/src/agents/plugins/pi/__tests__/pi.models.test.ts
index f734c6164..47829cc06 100644
--- a/src/agents/plugins/pi/__tests__/pi.models.test.ts
+++ b/src/agents/plugins/pi/__tests__/pi.models.test.ts
@@ -253,3 +253,35 @@ describe('convertLlmModelToPiEntry — cost', () => {
     expect(entry.compat).toEqual({ forceAdaptiveThinking: true });
   });
 });
+
+describe('convertLlmModelToPiEntry — GPT-6 limits and routing', () => {
+  beforeEach(() => {
+    vi.mocked(lookupPrice).mockReset();
+    vi.mocked(lookupPrice).mockReturnValue(null);
+  });
+
+  function gpt6Model(id: string): LlmModel {
+    return llmModel({ base_name: id, deployment_name: id, label: id });
+  }
+
+  it('should report a 1050000-token window for gpt-6-luna', () => {
+    const entry = convertLlmModelToPiEntry(gpt6Model('gpt-6-luna'));
+
+    expect(entry.contextWindow).toBe(1050000);
+    expect(entry.maxTokens).toBe(128000);
+  });
+
+  it('should report a 1050000-token window for a vendor-prefixed id', () => {
+    const entry = convertLlmModelToPiEntry(gpt6Model('openai.gpt-6-sol'));
+
+    expect(entry.contextWindow).toBe(1050000);
+    expect(entry.maxTokens).toBe(128000);
+  });
+
+  it('should route gpt-6-luna via openai-responses with reasoning', () => {
+    const entry = convertLlmModelToPiEntry(gpt6Model('gpt-6-luna'));
+
+    expect(entry.api).toBe('openai-responses');
+    expect(entry.reasoning).toBe(true);
+  });
+});
diff --git a/src/agents/plugins/pi/pi.models.ts b/src/agents/plugins/pi/pi.models.ts
index 8348de219..3ad9fa91f 100644
--- a/src/agents/plugins/pi/pi.models.ts
+++ b/src/agents/plugins/pi/pi.models.ts
@@ -24,6 +24,7 @@ const RESPONSES_API_PATTERNS: RegExp[] = [
   /^gpt-5\.5-/,
   /^gpt-5-5-/,
   /gpt-5[.-]6/,
+  /gpt-6[.-]/,
 ];
 
 export function classifyPiModel(modelId: string): PiModelClassification {
@@ -69,6 +70,7 @@ function detectLimits(id: string): { contextWindow: number; maxTokens: number }
   if (id.startsWith('gpt-4.1')) return { contextWindow: 1048576, maxTokens: 32768 };
   if (/^gpt-5\.5-/.test(id) || /^gpt-5-5-/.test(id)) return { contextWindow: 1050000, maxTokens: 128000 };
   if (/gpt-5[.-]6/.test(id)) return { contextWindow: 1050000, maxTokens: 128000 };
+  if (id.startsWith('gpt-6') || /gpt-6[.-]/.test(id)) return { contextWindow: 1050000, maxTokens: 128000 };
   if (id.startsWith('gpt-5')) return { contextWindow: 400000, maxTokens: 128000 };
   if (/^o[134]-/.test(id) || id === 'o1') return { contextWindow: 200000, maxTokens: 100000 };
   if (id.startsWith('qwen') || id.startsWith('moonshotai') || id.startsWith('kimi')) {
@@ -96,6 +98,8 @@ function isReasoningModel(id: string): boolean {
     id.startsWith('gemini') ||
     id.startsWith('gpt-5') ||
     /gpt-5[.-]6/.test(id) ||
+    id.startsWith('gpt-6') ||
+    /gpt-6[.-]/.test(id) ||
     /^o[134]-/.test(id) ||
     id === 'o1' ||
     id.startsWith('deepseek') ||
diff --git a/src/cli/commands/proxy/connectors/__tests__/vscode-models.test.ts b/src/cli/commands/proxy/connectors/__tests__/vscode-models.test.ts
index 00a4dc7ca..b864b1197 100644
--- a/src/cli/commands/proxy/connectors/__tests__/vscode-models.test.ts
+++ b/src/cli/commands/proxy/connectors/__tests__/vscode-models.test.ts
@@ -24,8 +24,17 @@ describe('findVsCodeCapabilityEntry', () => {
     expect(findVsCodeCapabilityEntry('openai.gpt-5.6-luna')?.family).toBe('gpt-5.6-luna');
   });
 
-  it('returns undefined for a model with no capability-table family', () => {
-    expect(findVsCodeCapabilityEntry('gpt-6-sol')).toBeUndefined();
+  it('returns the table entry for a GPT-6 model', () => {
+    expect(findVsCodeCapabilityEntry('gpt-6-sol')).toMatchObject({
+      family: 'gpt-6-sol',
+      apiType: 'responses',
+      maxInputTokens: 922000,
+      maxOutputTokens: 128000,
+    });
+  });
+
+  it('resolves a vendor-prefixed GPT-6 tenant id to its family entry', () => {
+    expect(findVsCodeCapabilityEntry('openai.gpt-6-luna')?.family).toBe('gpt-6-luna');
   });
 });
 
diff --git a/src/cli/commands/proxy/connectors/__tests__/vscode.test.ts b/src/cli/commands/proxy/connectors/__tests__/vscode.test.ts
index 85d537b09..c7d5e94fb 100644
--- a/src/cli/commands/proxy/connectors/__tests__/vscode.test.ts
+++ b/src/cli/commands/proxy/connectors/__tests__/vscode.test.ts
@@ -25,6 +25,8 @@ const EXPECTED_MODEL_IDS = [
   'gpt-5.6-luna-2026-07-09',
   'gpt-5.6-sol-2026-07-09',
   'gpt-5.6-terra-2026-07-09',
+  'gpt-6-luna',
+  'gpt-6-sol',
   'gemini-3-flash',
   'gemini-3.1-pro',
   'gemini-3.5-flash',
@@ -170,7 +172,7 @@ describe('writeVsCodeLanguageModelsConfigAtPath', () => {
     }
   });
 
-  it('renders stateless Responses reasoning capabilities for GPT-5.5 and GPT-5.6', async () => {
+  it('renders stateless Responses reasoning capabilities for GPT-5.5, GPT-5.6 and GPT-6', async () => {
     await writeVsCodeLanguageModelsConfigAtPath(configPath, 'http://127.0.0.1:4001', 'gw-key');
 
     const providers = await readProviders();
@@ -180,6 +182,8 @@ describe('writeVsCodeLanguageModelsConfigAtPath', () => {
       ['gpt-5.6-luna-2026-07-09', ['none', 'low', 'medium', 'high', 'xhigh', 'max']],
       ['gpt-5.6-sol-2026-07-09', ['none', 'low', 'medium', 'high', 'xhigh', 'max']],
       ['gpt-5.6-terra-2026-07-09', ['none', 'low', 'medium', 'high', 'xhigh', 'max']],
+      ['gpt-6-luna', ['none', 'low', 'medium', 'high', 'xhigh', 'max']],
+      ['gpt-6-sol', ['none', 'low', 'medium', 'high', 'xhigh', 'max']],
     ]);
 
     for (const [id, efforts] of expectedEfforts) {
@@ -404,24 +408,26 @@ describe('writeVsCodeLanguageModelsConfigAtPath', () => {
   describe('full tenant catalog', () => {
     const GPT_6_SOL = { base_name: 'gpt-6-sol', label: 'GPT-6 Sol' };
 
-    it('writes every enabled tenant model in catalog order, including unknown families', async () => {
-      mockCatalog([...EXPECTED_MODEL_IDS, GPT_6_SOL]);
+    it('writes every enabled tenant model in catalog order, including table-backed GPT-6 models', async () => {
+      mockCatalog(EXPECTED_MODEL_IDS);
 
       const result = await writeVsCodeLanguageModelsConfigAtPath(configPath, 'http://127.0.0.1:4001', 'gw-key');
 
       const providers = await readProviders();
       const models = providers[0].models as Array<Record<string, unknown>>;
-      expect(result.modelCount).toBe(EXPECTED_MODEL_IDS.length + 1);
-      expect(models.map(model => model.id)).toEqual([...EXPECTED_MODEL_IDS, 'gpt-6-sol']);
+      expect(result.modelCount).toBe(EXPECTED_MODEL_IDS.length);
+      expect(models.map(model => model.id)).toEqual(EXPECTED_MODEL_IDS);
       expect(models.find(model => model.id === 'gpt-6-sol')).toMatchObject({
-        name: 'GPT-6 Sol',
+        name: 'gpt-6-sol',
         apiType: 'responses',
         zeroDataRetentionEnabled: true,
         url: 'http://127.0.0.1:4001/v1/responses',
+        maxInputTokens: 922000,
+        maxOutputTokens: 128000,
       });
     });
 
-    it('keeps catalog order when an unknown model comes first', async () => {
+    it('keeps catalog order when a GPT-6 model comes first', async () => {
       mockCatalog([GPT_6_SOL, 'claude-sonnet-5', 'gpt-4.1']);
 
       await writeVsCodeLanguageModelsConfigAtPath(configPath, 'http://127.0.0.1:4001', 'gw-key');
@@ -432,15 +438,15 @@ describe('writeVsCodeLanguageModelsConfigAtPath', () => {
     });
 
     it('rebuilds the list from the live catalog on every write, picking up new models', async () => {
-      mockCatalog([...EXPECTED_MODEL_IDS, GPT_6_SOL]);
+      mockCatalog(EXPECTED_MODEL_IDS);
       await writeVsCodeLanguageModelsConfigAtPath(configPath, 'http://127.0.0.1:4001', 'gw-key');
 
-      mockCatalog([...EXPECTED_MODEL_IDS, GPT_6_SOL, 'claude-sonnet-6']);
+      mockCatalog([...EXPECTED_MODEL_IDS, 'claude-sonnet-6']);
       const result = await writeVsCodeLanguageModelsConfigAtPath(configPath, 'http://127.0.0.1:4001', 'gw-key');
 
       const providers = await readProviders();
       const models = providers[0].models as Array<Record<string, unknown>>;
-      expect(result.modelCount).toBe(EXPECTED_MODEL_IDS.length + 2);
+      expect(result.modelCount).toBe(EXPECTED_MODEL_IDS.length + 1);
       expect(models.map(model => model.id)).toContain('claude-sonnet-6');
     });
 
diff --git a/src/cli/commands/proxy/connectors/vscode-models.ts b/src/cli/commands/proxy/connectors/vscode-models.ts
index 6c7c45020..485393d3e 100644
--- a/src/cli/commands/proxy/connectors/vscode-models.ts
+++ b/src/cli/commands/proxy/connectors/vscode-models.ts
@@ -169,6 +169,28 @@ export const VS_CODE_CAPABILITY_TABLE: readonly VsCodeCapabilityEntry[] = [
     maxInputTokens: 922000,
     maxOutputTokens: 128000,
   },
+  {
+    family: 'gpt-6-luna',
+    apiType: 'responses',
+    vision: true,
+    thinking: true,
+    zeroDataRetentionEnabled: true,
+    supportsReasoningEffort: GPT_5_6_EFFORTS,
+    reasoningEffortFormat: 'responses',
+    maxInputTokens: 922000,
+    maxOutputTokens: 128000,
+  },
+  {
+    family: 'gpt-6-sol',
+    apiType: 'responses',
+    vision: true,
+    thinking: true,
+    zeroDataRetentionEnabled: true,
+    supportsReasoningEffort: GPT_5_6_EFFORTS,
+    reasoningEffortFormat: 'responses',
+    maxInputTokens: 922000,
+    maxOutputTokens: 128000,
+  },
   {
     family: 'gemini-3-flash',
     apiType: 'chat-completions',

From 8417673e3208444b32ced569fbdcdf071f324184 Mon Sep 17 00:00:00 2001
From: Siarhei Yarkavy <siarhei.yarkavy@gmail.com>
Date: Mon, 28 Sep 2026 19:04:32 +0300
Subject: [PATCH 2/8] fix(proxy): use dedicated GPT-6 efforts for VS Code
 capabilities

Address review on #582: gpt-6-luna/sol now use GPT_6_EFFORTS instead of GPT_5_6_EFFORTS. Same values, isolated for future divergence.
---
 src/cli/commands/proxy/connectors/vscode-models.ts | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/src/cli/commands/proxy/connectors/vscode-models.ts b/src/cli/commands/proxy/connectors/vscode-models.ts
index 485393d3e..1348279f9 100644
--- a/src/cli/commands/proxy/connectors/vscode-models.ts
+++ b/src/cli/commands/proxy/connectors/vscode-models.ts
@@ -36,6 +36,7 @@ const GPT_5_EFFORTS = ['minimal', 'low', 'medium', 'high'] as const;
 const GPT_5_2_EFFORTS = ['none', 'low', 'medium', 'high'] as const;
 const GPT_5_XHIGH_EFFORTS = ['none', 'low', 'medium', 'high', 'xhigh'] as const;
 const GPT_5_6_EFFORTS = ['none', 'low', 'medium', 'high', 'xhigh', 'max'] as const;
+const GPT_6_EFFORTS = ['none', 'low', 'medium', 'high', 'xhigh', 'max'] as const;
 const GEMINI_FLASH_EFFORTS = ['minimal', 'low', 'medium', 'high'] as const;
 const GEMINI_PRO_EFFORTS = ['low', 'medium', 'high'] as const;
 const CLAUDE_EFFORTS = ['low', 'medium', 'high'] as const;
@@ -175,7 +176,7 @@ export const VS_CODE_CAPABILITY_TABLE: readonly VsCodeCapabilityEntry[] = [
     vision: true,
     thinking: true,
     zeroDataRetentionEnabled: true,
-    supportsReasoningEffort: GPT_5_6_EFFORTS,
+    supportsReasoningEffort: GPT_6_EFFORTS,
     reasoningEffortFormat: 'responses',
     maxInputTokens: 922000,
     maxOutputTokens: 128000,
@@ -186,7 +187,7 @@ export const VS_CODE_CAPABILITY_TABLE: readonly VsCodeCapabilityEntry[] = [
     vision: true,
     thinking: true,
     zeroDataRetentionEnabled: true,
-    supportsReasoningEffort: GPT_5_6_EFFORTS,
+    supportsReasoningEffort: GPT_6_EFFORTS,
     reasoningEffortFormat: 'responses',
     maxInputTokens: 922000,
     maxOutputTokens: 128000,

From 5482355e9ef66d3f8c5bfafa7c0998f59e78606e Mon Sep 17 00:00:00 2001
From: Siarhei Yarkavy <siarhei.yarkavy@gmail.com>
Date: Thu, 1 Oct 2026 12:59:34 +0300
Subject: [PATCH 3/8] test(proxy): cover chat and responses fallbacks for
 unknown models

---
 tests/integration/vscode-byok.test.ts | 22 ++++++++++++++++------
 1 file changed, 16 insertions(+), 6 deletions(-)

diff --git a/tests/integration/vscode-byok.test.ts b/tests/integration/vscode-byok.test.ts
index 765fe1d5e..4d0bf53d9 100644
--- a/tests/integration/vscode-byok.test.ts
+++ b/tests/integration/vscode-byok.test.ts
@@ -35,7 +35,12 @@ import {
 
 const GATEWAY_KEY = 'test-local-key';
 const PROFILE_MODEL = 'profile-selected-model-that-must-not-be-used';
-const UNKNOWN_MODEL = 'gpt-6-sol';
+// Neutral unknown: no gpt-/claude-/codex prefix, so buildDefaultVsCodeCapability
+// falls back to conservative chat-completions defaults.
+const UNKNOWN_CHAT_MODEL = 'totally-unknown-model';
+// Responses unknown: gpt-7 major guarantees Responses defaults via
+// isResponsesOnlyGpt, while the segment never matches a capability-table family.
+const UNKNOWN_RESPONSES_MODEL = 'gpt-7-future-unknown-xyz';
 
 interface StartedServer {
   server: Server;
@@ -208,11 +213,12 @@ describe('VS Code BYOK model matrix', () => {
       // serve it directly from every family in the capability table so each
       // entry resolves as an exact match — the model matrix below then
       // exercises the request-forwarding behavior, not the resolver itself.
-      // One extra id with no capability family must still be written.
+      // Two extra ids with no capability family must still be written:
+      // a neutral chat-completions fallback and a Responses fallback.
       if (req.url?.startsWith('/v1/llm_models')) {
         res.setHeader('content-type', 'application/json');
         res.end(JSON.stringify({
-          data: [...VS_CODE_CAPABILITY_TABLE.map(entry => ({ id: entry.family })), { id: UNKNOWN_MODEL }],
+          data: [...VS_CODE_CAPABILITY_TABLE.map(entry => ({ id: entry.family })), { id: UNKNOWN_CHAT_MODEL }, { id: UNKNOWN_RESPONSES_MODEL }],
         }));
         return;
       }
@@ -240,10 +246,14 @@ describe('VS Code BYOK model matrix', () => {
       provider => provider.name === 'CodeMie' && provider.vendor === 'customendpoint'
     );
 
-    expect(codeMieProvider?.models).toHaveLength(VS_CODE_CAPABILITY_TABLE.length + 1);
+    expect(codeMieProvider?.models).toHaveLength(VS_CODE_CAPABILITY_TABLE.length + 2);
     expect(codeMieProvider?.models?.some(model => model.id === PROFILE_MODEL)).toBe(false);
-    expect(codeMieProvider?.models?.find(model => model.id === UNKNOWN_MODEL)).toMatchObject({
-      id: UNKNOWN_MODEL,
+    expect(codeMieProvider?.models?.find(model => model.id === UNKNOWN_CHAT_MODEL)).toMatchObject({
+      id: UNKNOWN_CHAT_MODEL,
+      apiType: 'chat-completions',
+    });
+    expect(codeMieProvider?.models?.find(model => model.id === UNKNOWN_RESPONSES_MODEL)).toMatchObject({
+      id: UNKNOWN_RESPONSES_MODEL,
       apiType: 'responses',
       zeroDataRetentionEnabled: true,
     });

From bb3fe16fb4a38b115aa949600befa1143a42829f Mon Sep 17 00:00:00 2001
From: Siarhei Yarkavy <siarhei.yarkavy@gmail.com>
Date: Sun, 4 Oct 2026 21:20:53 +0300
Subject: [PATCH 4/8] fix(agents): match bare and future GPT-6 ids in OpenCode
 and Pi

Replace startsWith('gpt-6') || /gpt-6[.-]/ with /gpt-6(?!\d)/ so bare vendor-prefixed ids (openai.gpt-6) and dotted variants (gpt-6.1-sol, future 6.2+) route via Responses API with 1050000/128000 limits, while gpt-60-style ids stay out.
---
 src/agents/plugins/opencode/opencode-dynamic-models.ts | 8 ++++----
 src/agents/plugins/pi/pi.models.ts                     | 9 +++++----
 2 files changed, 9 insertions(+), 8 deletions(-)

diff --git a/src/agents/plugins/opencode/opencode-dynamic-models.ts b/src/agents/plugins/opencode/opencode-dynamic-models.ts
index 209065422..0b2b4aaa4 100644
--- a/src/agents/plugins/opencode/opencode-dynamic-models.ts
+++ b/src/agents/plugins/opencode/opencode-dynamic-models.ts
@@ -29,7 +29,7 @@ import { logger } from '../../../utils/logger.js';
 // Naming conventions observed in CodeMie deployments:
 //   Responses API  →  gpt-5-2-*, gpt-5.2-*, gpt-5.x-codex-*, gpt-5-x-codex-*,
 //                     gpt-5.4-*, gpt-5-4-*, gpt-5.5-*, gpt-5-5-*, gpt-5.6-*, gpt-5-6-*,
-//                     gpt-6-*, gpt-6.* (e.g. gpt-6-luna, openai.gpt-6-luna)
+//                     gpt-6-*, gpt-6.* (e.g. gpt-6-luna, gpt-6.1-sol, openai.gpt-6-luna, bare openai.gpt-6)
 //   Chat Completions → gpt-4*, gpt-5-<year>-*, o1/o3/o4*, gemini-*, claude-*, …
 //
 // Update this list whenever new Responses-API-only models are deployed.
@@ -46,7 +46,7 @@ const RESPONSES_API_MODEL_PATTERNS: RegExp[] = [
   /^gpt-5\.5-/,       // gpt-5.5-2026-04-24 — same Azure restriction as gpt-5.4
   /^gpt-5-5-/,        // hyphenated variant of gpt-5.5-*
   /gpt-5[.-]6/,       // gpt-5.6-* (e.g. openai.gpt-5.6-luna) — same Azure restriction (tools + reasoning_effort)
-  /gpt-6[.-]/,        // gpt-6-* (e.g. gpt-6-luna, openai.gpt-6-luna) — Responses API for built-in tools
+  /gpt-6(?!\d)/,      // gpt-6-* / gpt-6.* (e.g. gpt-6-luna, gpt-6.1-sol, openai.gpt-6-luna, chatgpt-6.1-sol) — Responses API for built-in tools; (?!\d) keeps gpt-60-style ids out while staying ready for 6.2+
 ];
 
 function isResponsesApiModel(id: string): boolean {
@@ -60,7 +60,7 @@ function detectFamily(id: string): string {
   if (id.startsWith('gemini')) return 'gemini-2';
   if (id.startsWith('gpt-4')) return 'gpt-4';
   if (id.startsWith('gpt-5') || /gpt-5[.-]6/.test(id)) return 'gpt-5';
-  if (id.startsWith('gpt-6') || /gpt-6[.-]/.test(id)) return 'gpt-6';
+  if (/gpt-6(?!\d)/.test(id)) return 'gpt-6';
   if (/^o[134]-/.test(id) || id === 'o1') return 'openai-reasoning';
   if (id.startsWith('qwen')) return 'qwen3';
   if (id.startsWith('deepseek')) return 'deepseek';
@@ -80,7 +80,7 @@ function detectLimits(id: string, family: string): { context: number; output: nu
   if (id.startsWith('gpt-4o')) return { context: 128000, output: 16384 };
   if (id.startsWith('gpt-5.5') || id.startsWith('gpt-5-5')) return { context: 1050000, output: 128000 }; // Azure-published window for gpt-5.5
   if (/gpt-5[.-]6/.test(id)) return { context: 1050000, output: 128000 }; // Azure-published window for gpt-5.6
-  if (id.startsWith('gpt-6') || /gpt-6[.-]/.test(id)) return { context: 1050000, output: 128000 }; // Vendor-published window for gpt-6 (1,050,000 context, 128k output)
+  if (/gpt-6(?!\d)/.test(id)) return { context: 1050000, output: 128000 }; // Vendor-published window for gpt-6 (1,050,000 context, 128k output; covers 6.1-sol, future 6.2+)
   if (id.startsWith('gpt-5')) return { context: 400000, output: 128000 };
   if (/^o[134]-/.test(id) || id === 'o1') return { context: 200000, output: 100000 };
   if (id.startsWith('qwen') || id.startsWith('moonshotai') || id.startsWith('kimi')) return { context: 262144, output: 131072 };
diff --git a/src/agents/plugins/pi/pi.models.ts b/src/agents/plugins/pi/pi.models.ts
index 3ad9fa91f..9a4a47e43 100644
--- a/src/agents/plugins/pi/pi.models.ts
+++ b/src/agents/plugins/pi/pi.models.ts
@@ -24,7 +24,9 @@ const RESPONSES_API_PATTERNS: RegExp[] = [
   /^gpt-5\.5-/,
   /^gpt-5-5-/,
   /gpt-5[.-]6/,
-  /gpt-6[.-]/,
+  // gpt-6-*/gpt-6.* (e.g. gpt-6-luna, gpt-6.1-sol, openai.gpt-6-luna, bare openai.gpt-6).
+  // (?!\d) keeps gpt-60-style ids out while staying ready for 6.2+.
+  /gpt-6(?!\d)/,
 ];
 
 export function classifyPiModel(modelId: string): PiModelClassification {
@@ -70,7 +72,7 @@ function detectLimits(id: string): { contextWindow: number; maxTokens: number }
   if (id.startsWith('gpt-4.1')) return { contextWindow: 1048576, maxTokens: 32768 };
   if (/^gpt-5\.5-/.test(id) || /^gpt-5-5-/.test(id)) return { contextWindow: 1050000, maxTokens: 128000 };
   if (/gpt-5[.-]6/.test(id)) return { contextWindow: 1050000, maxTokens: 128000 };
-  if (id.startsWith('gpt-6') || /gpt-6[.-]/.test(id)) return { contextWindow: 1050000, maxTokens: 128000 };
+  if (/gpt-6(?!\d)/.test(id)) return { contextWindow: 1050000, maxTokens: 128000 };
   if (id.startsWith('gpt-5')) return { contextWindow: 400000, maxTokens: 128000 };
   if (/^o[134]-/.test(id) || id === 'o1') return { contextWindow: 200000, maxTokens: 100000 };
   if (id.startsWith('qwen') || id.startsWith('moonshotai') || id.startsWith('kimi')) {
@@ -98,8 +100,7 @@ function isReasoningModel(id: string): boolean {
     id.startsWith('gemini') ||
     id.startsWith('gpt-5') ||
     /gpt-5[.-]6/.test(id) ||
-    id.startsWith('gpt-6') ||
-    /gpt-6[.-]/.test(id) ||
+    /gpt-6(?!\d)/.test(id) ||
     /^o[134]-/.test(id) ||
     id === 'o1' ||
     id.startsWith('deepseek') ||

From ef3bca676a363c4b3ad9d245106828382490b8d2 Mon Sep 17 00:00:00 2001
From: Siarhei Yarkavy <siarhei.yarkavy@gmail.com>
Date: Mon, 5 Oct 2026 17:11:39 +0300
Subject: [PATCH 5/8] fix: add gpt-6-astra to VS Code capabilities and OpenCode
 configs

VS Code capability table: gpt-6-astra entry (922000/128000, Responses, GPT_6_EFFORTS) alongside luna/sol.

Addresses review thread r4183491171 in #582.
---
 .../opencode/opencode-model-configs.ts        | 29 +++++++++++++++++++
 .../proxy/connectors/__tests__/vscode.test.ts |  2 ++
 .../proxy/connectors/vscode-models.ts         | 11 +++++++
 3 files changed, 42 insertions(+)

diff --git a/src/agents/plugins/opencode/opencode-model-configs.ts b/src/agents/plugins/opencode/opencode-model-configs.ts
index e87ac3418..85068943d 100644
--- a/src/agents/plugins/opencode/opencode-model-configs.ts
+++ b/src/agents/plugins/opencode/opencode-model-configs.ts
@@ -347,6 +347,35 @@ export const OPENCODE_MODEL_CONFIGS: Record<string, OpenCodeModelConfig> = {
       output: 128000
     }
   },
+  'gpt-6-astra': {
+    id: 'gpt-6-astra',
+    name: 'GPT-6 Astra',
+    displayName: 'GPT-6 Astra',
+    family: 'gpt-6',
+    tool_call: true,
+    reasoning: true,
+    attachment: true,
+    temperature: false,
+    structured_output: true,
+    use_responses_api: true,
+    modalities: {
+      input: ['text', 'image'],
+      output: ['text']
+    },
+    knowledge: '2026-05-18',
+    release_date: '2026-09-22',
+    last_updated: '2026-09-22',
+    open_weights: false,
+    cost: {
+      input: 10,
+      output: 50,
+      cache_read: 1
+    },
+    limit: {
+      context: 1050000,
+      output: 128000
+    }
+  },
 
   // ── Claude Models ──────────────────────────────────────────────────
   'claude-4-5-sonnet': {
diff --git a/src/cli/commands/proxy/connectors/__tests__/vscode.test.ts b/src/cli/commands/proxy/connectors/__tests__/vscode.test.ts
index c7d5e94fb..ff551a3ef 100644
--- a/src/cli/commands/proxy/connectors/__tests__/vscode.test.ts
+++ b/src/cli/commands/proxy/connectors/__tests__/vscode.test.ts
@@ -27,6 +27,7 @@ const EXPECTED_MODEL_IDS = [
   'gpt-5.6-terra-2026-07-09',
   'gpt-6-luna',
   'gpt-6-sol',
+  'gpt-6-astra',
   'gemini-3-flash',
   'gemini-3.1-pro',
   'gemini-3.5-flash',
@@ -184,6 +185,7 @@ describe('writeVsCodeLanguageModelsConfigAtPath', () => {
       ['gpt-5.6-terra-2026-07-09', ['none', 'low', 'medium', 'high', 'xhigh', 'max']],
       ['gpt-6-luna', ['none', 'low', 'medium', 'high', 'xhigh', 'max']],
       ['gpt-6-sol', ['none', 'low', 'medium', 'high', 'xhigh', 'max']],
+      ['gpt-6-astra', ['none', 'low', 'medium', 'high', 'xhigh', 'max']],
     ]);
 
     for (const [id, efforts] of expectedEfforts) {
diff --git a/src/cli/commands/proxy/connectors/vscode-models.ts b/src/cli/commands/proxy/connectors/vscode-models.ts
index 1348279f9..9ae86bf61 100644
--- a/src/cli/commands/proxy/connectors/vscode-models.ts
+++ b/src/cli/commands/proxy/connectors/vscode-models.ts
@@ -192,6 +192,17 @@ export const VS_CODE_CAPABILITY_TABLE: readonly VsCodeCapabilityEntry[] = [
     maxInputTokens: 922000,
     maxOutputTokens: 128000,
   },
+  {
+    family: 'gpt-6-astra',
+    apiType: 'responses',
+    vision: true,
+    thinking: true,
+    zeroDataRetentionEnabled: true,
+    supportsReasoningEffort: GPT_6_EFFORTS,
+    reasoningEffortFormat: 'responses',
+    maxInputTokens: 922000,
+    maxOutputTokens: 128000,
+  },
   {
     family: 'gemini-3-flash',
     apiType: 'chat-completions',

From c751f5c8a9a779bf66e490b26477fdbd0168d84a Mon Sep 17 00:00:00 2001
From: Siarhei Yarkavy <siarhei.yarkavy@gmail.com>
Date: Mon, 5 Oct 2026 17:19:39 +0300
Subject: [PATCH 6/8] fix: add gpt-6.1-sol to VS Code capabilities and OpenCode
 configs

VS Code capability table: gpt-6.1-sol entry (922000/128000, Responses, GPT_6_EFFORTS); dotted family key per gpt-5.6 precedent, resolves dashed, vendor-prefixed and dated ids.

Static OpenCode config: 1050000/128000; cost and dates mirrored from gpt-6-sol pending vendor pricing verification (no pricing row for 6.1-sol yet; no 6.1-astra shipped).
---
 .../opencode/opencode-model-configs.ts        | 29 +++++++++++++++++++
 .../proxy/connectors/__tests__/vscode.test.ts |  2 ++
 .../proxy/connectors/vscode-models.ts         | 11 +++++++
 3 files changed, 42 insertions(+)

diff --git a/src/agents/plugins/opencode/opencode-model-configs.ts b/src/agents/plugins/opencode/opencode-model-configs.ts
index 85068943d..293fec55b 100644
--- a/src/agents/plugins/opencode/opencode-model-configs.ts
+++ b/src/agents/plugins/opencode/opencode-model-configs.ts
@@ -376,6 +376,35 @@ export const OPENCODE_MODEL_CONFIGS: Record<string, OpenCodeModelConfig> = {
       output: 128000
     }
   },
+  'gpt-6.1-sol': {
+    id: 'gpt-6.1-sol',
+    name: 'GPT-6.1 Sol',
+    displayName: 'GPT-6.1 Sol',
+    family: 'gpt-6',
+    tool_call: true,
+    reasoning: true,
+    attachment: true,
+    temperature: false,
+    structured_output: true,
+    use_responses_api: true,
+    modalities: {
+      input: ['text', 'image'],
+      output: ['text']
+    },
+    knowledge: '2026-05-18',
+    release_date: '2026-09-22',
+    last_updated: '2026-09-22',
+    open_weights: false,
+    cost: {
+      input: 2,
+      output: 10,
+      cache_read: 0.20
+    },
+    limit: {
+      context: 1050000,
+      output: 128000
+    }
+  },
 
   // ── Claude Models ──────────────────────────────────────────────────
   'claude-4-5-sonnet': {
diff --git a/src/cli/commands/proxy/connectors/__tests__/vscode.test.ts b/src/cli/commands/proxy/connectors/__tests__/vscode.test.ts
index ff551a3ef..5e3256d80 100644
--- a/src/cli/commands/proxy/connectors/__tests__/vscode.test.ts
+++ b/src/cli/commands/proxy/connectors/__tests__/vscode.test.ts
@@ -28,6 +28,7 @@ const EXPECTED_MODEL_IDS = [
   'gpt-6-luna',
   'gpt-6-sol',
   'gpt-6-astra',
+  'gpt-6.1-sol',
   'gemini-3-flash',
   'gemini-3.1-pro',
   'gemini-3.5-flash',
@@ -186,6 +187,7 @@ describe('writeVsCodeLanguageModelsConfigAtPath', () => {
       ['gpt-6-luna', ['none', 'low', 'medium', 'high', 'xhigh', 'max']],
       ['gpt-6-sol', ['none', 'low', 'medium', 'high', 'xhigh', 'max']],
       ['gpt-6-astra', ['none', 'low', 'medium', 'high', 'xhigh', 'max']],
+      ['gpt-6.1-sol', ['none', 'low', 'medium', 'high', 'xhigh', 'max']],
     ]);
 
     for (const [id, efforts] of expectedEfforts) {
diff --git a/src/cli/commands/proxy/connectors/vscode-models.ts b/src/cli/commands/proxy/connectors/vscode-models.ts
index 9ae86bf61..a794e23b8 100644
--- a/src/cli/commands/proxy/connectors/vscode-models.ts
+++ b/src/cli/commands/proxy/connectors/vscode-models.ts
@@ -203,6 +203,17 @@ export const VS_CODE_CAPABILITY_TABLE: readonly VsCodeCapabilityEntry[] = [
     maxInputTokens: 922000,
     maxOutputTokens: 128000,
   },
+  {
+    family: 'gpt-6.1-sol',
+    apiType: 'responses',
+    vision: true,
+    thinking: true,
+    zeroDataRetentionEnabled: true,
+    supportsReasoningEffort: GPT_6_EFFORTS,
+    reasoningEffortFormat: 'responses',
+    maxInputTokens: 922000,
+    maxOutputTokens: 128000,
+  },
   {
     family: 'gemini-3-flash',
     apiType: 'chat-completions',

From b7892f93d81de457d7b083eb4e98fef04aedef2e Mon Sep 17 00:00:00 2001
From: Siarhei Yarkavy <siarhei.yarkavy@gmail.com>
Date: Mon, 5 Oct 2026 17:29:28 +0300
Subject: [PATCH 7/8] fix: correct gpt-6.1-sol pricing from vendor Standard
 table

Static OpenCode config cache_read 0.20 -> 0.1; add gpt-6-1-sol row to pricing.json (2/10/0.1/2.5, dashed key per file convention; dotted and dated ids resolve via normalizeModelName/snapshot).

Source: https://developers.openai.com/api/docs/pricing?latest-pricing=standard (Standard pricing table, row 'gpt-6.1-sol'). Supersedes the mirrored-from-gpt-6-sol placeholder in c751f5c8.
---
 src/agents/plugins/opencode/opencode-model-configs.ts | 2 +-
 src/utils/pricing.json                                | 7 +++++++
 2 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/src/agents/plugins/opencode/opencode-model-configs.ts b/src/agents/plugins/opencode/opencode-model-configs.ts
index 293fec55b..af5992bc6 100644
--- a/src/agents/plugins/opencode/opencode-model-configs.ts
+++ b/src/agents/plugins/opencode/opencode-model-configs.ts
@@ -398,7 +398,7 @@ export const OPENCODE_MODEL_CONFIGS: Record<string, OpenCodeModelConfig> = {
     cost: {
       input: 2,
       output: 10,
-      cache_read: 0.20
+      cache_read: 0.1
     },
     limit: {
       context: 1050000,
diff --git a/src/utils/pricing.json b/src/utils/pricing.json
index bac8eb008..c46f8283d 100644
--- a/src/utils/pricing.json
+++ b/src/utils/pricing.json
@@ -16,6 +16,7 @@
       "keys": "All keys use dashes instead of dots (e.g. gpt-5-4 not gpt-5.4) so normalizeModelName dots→dashes works",
       "cacheRead": "cache hit price (Anthropic: 0.1x input, OpenAI: varies, Google: ~0.1x input)",
       "cacheWrite": "cache write price (Anthropic: 1.25x input, OpenAI: ~same as input)",
+      "gpt61Sol": "Row added 2026-10-05: gpt-6-1-sol short-context Standard $2.00/$0.10/$2.50/$10.00 verified against https://developers.openai.com/api/docs/pricing?latest-pricing=standard (Standard pricing table, row 'gpt-6.1-sol'). Dashed key per this file's convention; dotted ids normalize via normalizeModelName.",
       "claude5": "Sonnet 5 and Opus 5 standard API-equivalent token rates verified 2026-09-15; Fable 5/5.1, Mythos 5/5.1 and Opus 4.7/4.8 verified 2026-09-27 at https://platform.claude.com/docs/en/about-claude/pricing; excludes additional tool charges. Sonnet 5's planned September 1 price increase was cancelled.",
       "t10CatalogReconciliation": "Rows added 2026-09-26 to price catalog ids that resolvePrice() found unpriced (unify-analytics-cost-command T10), each verified against one official vendor page/API on that date. claude-opus-5-5: $4/$20 base, $5/$8 5m/1h cache write, $0.20 cache read (0.05x multiplier), bedrockRegionalMultiplier 1.1 (Opus 4.5+ scope) — https://platform.claude.com/docs/en/about-claude/pricing (Model pricing table row 'Claude Opus 5.5'). gpt-5-5: short-context Standard row '$5.00 | $0.50 | - | $30.00' — https://developers.openai.com/api/docs/pricing ('Standard Pricing Tables', row 'gpt-5.5 (<272K context length)'); cache-write column is '-' (not stated) so cacheWrite mirrors input, matching this table's existing gpt-5-4 handling. gpt-5-6-luna/-sol/-terra and gpt-6-luna/-sol: short-context Standard rows from the same OpenAI page/table, e.g. 'gpt-6-sol | $2.00 | $0.20 | $2.50 | $10.00'; these rows do state a real cache-write price (~1.25x input) so it is recorded as-is rather than mirrored. deepseek-v4-pro: peak (weekday 01:00-04:00/06:00-10:00 UTC) cache-miss input $1.32, cache-hit input $0.044, output $3.96 — https://api-docs.deepseek.com/quick_start/pricing/; off-peak (all other hours, including weekends) is exactly half: $0.66/$0.022/$1.98; peak chosen as the flat row like every other entry in this table, no time-of-day tiering supported by the schema; cacheWrite mirrors input per this table's deepseek-chat/deepseek-reasoner precedent. gemini-3-6-flash/gemini-3-8-flash: paid-tier Standard pricing 'through Dec 31, 2026' — $0.75 input/$3.75 output/$0.075 context-caching — https://ai.google.dev/gemini-api/docs/pricing; scheduled to double to $1.50/$7.50/$0.15 on 2027-01-01, not yet in effect as of verification; cacheWrite mirrors input per this table's gemini-3-5-flash/gemini-3-7-flash precedent. gemini-3-1-flash-image: Standard pricing 'Input: $0.50 (text/image); Output: $3 (text/thinking), $60.00 (images)' — https://ai.google.dev/gemini-api/docs/pricing; output field records the $60 image-generation rate (not the $3 text/thinking rate), matching this table's existing gemini-3-1-flash-image-preview/gemini-2-5-flash-image rows, which use the same convention (schema has no separate image-output field); cacheRead/cacheWrite 0 (no cache pricing published for this model). grok-4-6: base tier (<200k prompt tokens) row 'grok-4.6 (< 200k prompt tokens) | 500k | $2.00 | $0.50 | $6.00' — https://docs.x.ai/developers/models; xAI doubles every rate for prompts >=200k tokens, not represented here (schema is flat), so this is the base/short-context tier, consistent with this table's other short-context-only entries; cacheWrite mirrors input per this table's grok-4/grok-4-1-fast precedent. qwen3-coder-30b-a3b-v1 and qwen3-coder-480b-a35b-v1: AWS Bedrock on-demand Standard tier, US East (N. Virginia) — official AWS Price List API https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/20260926004940/us-east-1/index.json (also rendered at https://aws.amazon.com/bedrock/pricing/), SKU usagetype 'qwen.qwen3-coder-30b-a3b-instruct-mantle-{input,output}-tokens-standard' = $0.00015/$0.0006 per 1K tokens ($0.15/$0.60 per 1M) and 'qwen.qwen3-coder-480b-a35b-instruct-mantle-{input,output}-tokens-standard' = $0.00045/$0.0018 per 1K tokens ($0.45/$1.80 per 1M); no cache pricing published for either. Checked but NOT added (no official page states a price for the exact catalog id): claude-4-5-sonnet / claude-4-5-sonnet-vertex now resolve via the T1 reorder fallback onto the existing claude-sonnet-4-5 row, so no new row was needed."
     },
@@ -267,6 +268,12 @@
     "cacheRead": 0.2,
     "cacheWrite": 2.5
   },
+  "gpt-6-1-sol": {
+    "input": 2,
+    "output": 10,
+    "cacheRead": 0.1,
+    "cacheWrite": 2.5
+  },
   "gpt-5-2": {
     "input": 1.75,
     "output": 14,

From d08fb4eed1ee3f04f87ff1f43d6206c3406791c9 Mon Sep 17 00:00:00 2001
From: Siarhei Yarkavy <siarhei.yarkavy@gmail.com>
Date: Mon, 5 Oct 2026 17:31:24 +0300
Subject: [PATCH 8/8] fix: correct gpt-6-astra cacheWrite from vendor Standard
 table

cacheWrite 10 -> 12.5 per https://developers.openai.com/api/docs/pricing?latest-pricing=standard (Standard pricing table, row 'gpt-6-astra': short-context 0.00/.00/2.50/0.00). 10 was a mirrored-input error carried over from #583; real price is ~1.25x input like the other GPT-6 rows.
---
 src/utils/pricing.json | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/src/utils/pricing.json b/src/utils/pricing.json
index c46f8283d..30501037f 100644
--- a/src/utils/pricing.json
+++ b/src/utils/pricing.json
@@ -17,6 +17,7 @@
       "cacheRead": "cache hit price (Anthropic: 0.1x input, OpenAI: varies, Google: ~0.1x input)",
       "cacheWrite": "cache write price (Anthropic: 1.25x input, OpenAI: ~same as input)",
       "gpt61Sol": "Row added 2026-10-05: gpt-6-1-sol short-context Standard $2.00/$0.10/$2.50/$10.00 verified against https://developers.openai.com/api/docs/pricing?latest-pricing=standard (Standard pricing table, row 'gpt-6.1-sol'). Dashed key per this file's convention; dotted ids normalize via normalizeModelName.",
+      "astraCacheWriteFix": "Corrected 2026-10-05: gpt-6-astra cacheWrite 10 -> 12.5 per the same vendor Standard table (row 'gpt-6-astra': $10.00/$1.00/$12.50/$50.00); 10 was a mirrored-input error, real price is ~1.25x input.",
       "claude5": "Sonnet 5 and Opus 5 standard API-equivalent token rates verified 2026-09-15; Fable 5/5.1, Mythos 5/5.1 and Opus 4.7/4.8 verified 2026-09-27 at https://platform.claude.com/docs/en/about-claude/pricing; excludes additional tool charges. Sonnet 5's planned September 1 price increase was cancelled.",
       "t10CatalogReconciliation": "Rows added 2026-09-26 to price catalog ids that resolvePrice() found unpriced (unify-analytics-cost-command T10), each verified against one official vendor page/API on that date. claude-opus-5-5: $4/$20 base, $5/$8 5m/1h cache write, $0.20 cache read (0.05x multiplier), bedrockRegionalMultiplier 1.1 (Opus 4.5+ scope) — https://platform.claude.com/docs/en/about-claude/pricing (Model pricing table row 'Claude Opus 5.5'). gpt-5-5: short-context Standard row '$5.00 | $0.50 | - | $30.00' — https://developers.openai.com/api/docs/pricing ('Standard Pricing Tables', row 'gpt-5.5 (<272K context length)'); cache-write column is '-' (not stated) so cacheWrite mirrors input, matching this table's existing gpt-5-4 handling. gpt-5-6-luna/-sol/-terra and gpt-6-luna/-sol: short-context Standard rows from the same OpenAI page/table, e.g. 'gpt-6-sol | $2.00 | $0.20 | $2.50 | $10.00'; these rows do state a real cache-write price (~1.25x input) so it is recorded as-is rather than mirrored. deepseek-v4-pro: peak (weekday 01:00-04:00/06:00-10:00 UTC) cache-miss input $1.32, cache-hit input $0.044, output $3.96 — https://api-docs.deepseek.com/quick_start/pricing/; off-peak (all other hours, including weekends) is exactly half: $0.66/$0.022/$1.98; peak chosen as the flat row like every other entry in this table, no time-of-day tiering supported by the schema; cacheWrite mirrors input per this table's deepseek-chat/deepseek-reasoner precedent. gemini-3-6-flash/gemini-3-8-flash: paid-tier Standard pricing 'through Dec 31, 2026' — $0.75 input/$3.75 output/$0.075 context-caching — https://ai.google.dev/gemini-api/docs/pricing; scheduled to double to $1.50/$7.50/$0.15 on 2027-01-01, not yet in effect as of verification; cacheWrite mirrors input per this table's gemini-3-5-flash/gemini-3-7-flash precedent. gemini-3-1-flash-image: Standard pricing 'Input: $0.50 (text/image); Output: $3 (text/thinking), $60.00 (images)' — https://ai.google.dev/gemini-api/docs/pricing; output field records the $60 image-generation rate (not the $3 text/thinking rate), matching this table's existing gemini-3-1-flash-image-preview/gemini-2-5-flash-image rows, which use the same convention (schema has no separate image-output field); cacheRead/cacheWrite 0 (no cache pricing published for this model). grok-4-6: base tier (<200k prompt tokens) row 'grok-4.6 (< 200k prompt tokens) | 500k | $2.00 | $0.50 | $6.00' — https://docs.x.ai/developers/models; xAI doubles every rate for prompts >=200k tokens, not represented here (schema is flat), so this is the base/short-context tier, consistent with this table's other short-context-only entries; cacheWrite mirrors input per this table's grok-4/grok-4-1-fast precedent. qwen3-coder-30b-a3b-v1 and qwen3-coder-480b-a35b-v1: AWS Bedrock on-demand Standard tier, US East (N. Virginia) — official AWS Price List API https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrock/20260926004940/us-east-1/index.json (also rendered at https://aws.amazon.com/bedrock/pricing/), SKU usagetype 'qwen.qwen3-coder-30b-a3b-instruct-mantle-{input,output}-tokens-standard' = $0.00015/$0.0006 per 1K tokens ($0.15/$0.60 per 1M) and 'qwen.qwen3-coder-480b-a35b-instruct-mantle-{input,output}-tokens-standard' = $0.00045/$0.0018 per 1K tokens ($0.45/$1.80 per 1M); no cache pricing published for either. Checked but NOT added (no official page states a price for the exact catalog id): claude-4-5-sonnet / claude-4-5-sonnet-vertex now resolve via the T1 reorder fallback onto the existing claude-sonnet-4-5 row, so no new row was needed."
     },
@@ -242,7 +243,7 @@
     "input": 10,
     "output": 50,
     "cacheRead": 1,
-    "cacheWrite": 10
+    "cacheWrite": 12.5
   },
   "gpt-5-5-pro": {
     "input": 30,