agentic-qe 3.14.3 → 3.14.5

package/dist/shared/llm/anthropic-params.d.ts

@@ -0,0 +1,40 @@
+/**
+ * Agentic QE v3 - Anthropic Messages API parameter policy
+ *
+ * Newer Claude models reject request fields that older ones accepted:
+ * `temperature` / `top_p` / `top_k` return 400 ("`temperature` is deprecated
+ * for this model") on Opus 4.7+, Sonnet 5 and the Fable family, and effort
+ * must be sent as `output_config.effort` (a nested `thinking.effort` is
+ * rejected). Every Anthropic-format request builder applies this policy so a
+ * model bump cannot silently break calls.
+ */
+import { type EffortLevel } from './effort-resolver';
+export interface AnthropicParamPolicy {
+    /** Whether `temperature` (and other sampling params) may be sent. */
+    sendSampling: boolean;
+    /** Whether `output_config.effort` is accepted. */
+    supportsEffort: boolean;
+}
+/**
+ * Parameter policy for a model ID (canonical, dated, Bedrock, or OpenRouter
+ * form). Models missing from the registry get the conservative policy:
+ * no sampling params (current models reject them; retired models no longer
+ * serve requests) and no effort.
+ */
+export declare function getAnthropicParamPolicy(model: string): AnthropicParamPolicy;
+export interface AnthropicParamInput {
+    temperature?: number;
+    /**
+     * Explicit effort. When omitted, flagship (Opus/Fable-class) models get the
+     * fleet effort (QE_EFFORT_LEVEL > config/fleet-defaults.yaml > xhigh, per
+     * ADR-093); other models keep their API default to avoid raising cost on
+     * every Sonnet/Haiku call.
+     */
+    effort?: EffortLevel;
+}
+/**
+ * Add sampling and effort fields to an Anthropic Messages request body
+ * according to the model's policy. Mutates and returns `body`.
+ */
+export declare function applyAnthropicParams(body: Record<string, unknown>, model: string, input: AnthropicParamInput): Record<string, unknown>;
+//# sourceMappingURL=anthropic-params.d.ts.map

package/dist/shared/llm/anthropic-params.js

@@ -0,0 +1,51 @@
+/**
+ * Agentic QE v3 - Anthropic Messages API parameter policy
+ *
+ * Newer Claude models reject request fields that older ones accepted:
+ * `temperature` / `top_p` / `top_k` return 400 ("`temperature` is deprecated
+ * for this model") on Opus 4.7+, Sonnet 5 and the Fable family, and effort
+ * must be sent as `output_config.effort` (a nested `thinking.effort` is
+ * rejected). Every Anthropic-format request builder applies this policy so a
+ * model bump cannot silently break calls.
+ */
+import { getModelCapabilities, getModelInfo } from './model-registry.js';
+import { resolveEffortLevel } from './effort-resolver.js';
+/**
+ * Parameter policy for a model ID (canonical, dated, Bedrock, or OpenRouter
+ * form). Models missing from the registry get the conservative policy:
+ * no sampling params (current models reject them; retired models no longer
+ * serve requests) and no effort.
+ */
+export function getAnthropicParamPolicy(model) {
+    try {
+        const caps = getModelCapabilities(model);
+        return {
+            sendSampling: caps.supportsSamplingParams !== false,
+            supportsEffort: caps.supportsEffortXHigh === true,
+        };
+    }
+    catch {
+        return { sendSampling: false, supportsEffort: false };
+    }
+}
+/**
+ * Add sampling and effort fields to an Anthropic Messages request body
+ * according to the model's policy. Mutates and returns `body`.
+ */
+export function applyAnthropicParams(body, model, input) {
+    const policy = getAnthropicParamPolicy(model);
+    if (policy.sendSampling && input.temperature !== undefined) {
+        body.temperature = input.temperature;
+    }
+    if (policy.supportsEffort) {
+        const effort = input.effort ?? (isFlagship(model) ? resolveEffortLevel() : undefined);
+        if (effort !== undefined) {
+            body.output_config = { effort };
+        }
+    }
+    return body;
+}
+function isFlagship(model) {
+    return getModelInfo(model)?.tier === 'flagship';
+}
+//# sourceMappingURL=anthropic-params.js.map

package/dist/shared/llm/cost-tracker.d.ts

@@ -7,14 +7,27 @@
  */
 import { LLMProviderType, TokenUsage, CostInfo, CostPeriod, CostSummary } from './interfaces';
 /**
- * Token pricing per model (cost per 1M tokens in USD)
- * Prices as of early 2025 - should be updated periodically
+ * Token pricing per model (cost per 1M tokens in USD).
+ * Current-generation prices verified 2026-09-27 (Anthropic docs, OpenRouter
+ * catalog). Models missing here fall back to the model registry via
+ * `resolveModelPricing()` before being treated as free/local.
  */
 export declare const MODEL_PRICING: Record<string, {
     input: number;
     output: number;
     provider: LLMProviderType;
 }>;
+/**
+ * Resolve per-million-token pricing for a model. Exact `MODEL_PRICING` keys
+ * win; otherwise the model registry is consulted (it normalizes provider
+ * IDs such as `anthropic/claude-sonnet-5` or dated Anthropic IDs). Returns
+ * `undefined` only when neither source knows the model, which callers treat
+ * as a free/local model.
+ */
+export declare function resolveModelPricing(model: string): {
+    input: number;
+    output: number;
+} | undefined;
 /**
  * Usage record for a single request
  */

package/dist/shared/llm/cost-tracker.js

@@ -6,25 +6,41 @@
  * Supports cost limits, alerts, and detailed breakdowns.
  */
 import { randomUUID } from 'crypto';
+import { getModelCost } from './model-registry.js';
 /**
- * Token pricing per model (cost per 1M tokens in USD)
- * Prices as of early 2025 - should be updated periodically
+ * Token pricing per model (cost per 1M tokens in USD).
+ * Current-generation prices verified 2026-09-27 (Anthropic docs, OpenRouter
+ * catalog). Models missing here fall back to the model registry via
+ * `resolveModelPricing()` before being treated as free/local.
  */
 export const MODEL_PRICING = {
-    // Claude models
+    // Claude models — current generation (2026-09 model refresh)
+    'claude-sonnet-5': { input: 2.0, output: 10.0, provider: 'claude' },
+    'claude-opus-5': { input: 5.0, output: 25.0, provider: 'claude' },
+    'claude-opus-5-5': { input: 4.0, output: 20.0, provider: 'claude' },
+    'claude-fable-5-1': { input: 10.0, output: 50.0, provider: 'claude' },
+    'claude-fable-5': { input: 10.0, output: 50.0, provider: 'claude' },
+    'claude-opus-4-8': { input: 5.0, output: 25.0, provider: 'claude' },
     // ADR-093: Opus 4.7 (2026-04-16), Sonnet 4.6, Haiku 4.5 (Apr 2026 cluster)
     'claude-opus-4-7': { input: 5.0, output: 25.0, provider: 'claude' },
     'claude-sonnet-4-6': { input: 3.0, output: 15.0, provider: 'claude' },
     'claude-haiku-4-5-20251001': { input: 1.0, output: 5.0, provider: 'claude' },
-    // Retiring 2026-06-15 — kept for existing traffic during transition
-    'claude-opus-4-5-20251101': { input: 15.0, output: 75.0, provider: 'claude' },
+    'claude-haiku-4-5': { input: 1.0, output: 5.0, provider: 'claude' },
+    // Previous generation, still served
+    'claude-opus-4-5-20251101': { input: 5.0, output: 25.0, provider: 'claude' },
+    'claude-opus-4-6': { input: 5.0, output: 25.0, provider: 'claude' },
+    // Retired — kept only to price historical usage records
     'claude-sonnet-4-20250514': { input: 3.0, output: 15.0, provider: 'claude' },
     'claude-3-5-haiku-20241022': { input: 1.0, output: 5.0, provider: 'claude' },
-    // Legacy Claude models
+    // Legacy Claude models (retired)
     'claude-3-opus-20240229': { input: 15.0, output: 75.0, provider: 'claude' },
     'claude-3-sonnet-20240229': { input: 3.0, output: 15.0, provider: 'claude' },
     'claude-3-haiku-20240307': { input: 0.25, output: 1.25, provider: 'claude' },
-    // OpenAI models
+    // OpenAI models — GPT-6 generation (OpenRouter pass-through pricing)
+    'gpt-6-astra': { input: 10.0, output: 50.0, provider: 'openai' },
+    'gpt-6-sol': { input: 2.0, output: 10.0, provider: 'openai' },
+    'gpt-6-luna': { input: 0.1, output: 0.5, provider: 'openai' },
+    // OpenAI legacy generation
     'gpt-4o': { input: 5.0, output: 15.0, provider: 'openai' },
     'gpt-4o-mini': { input: 0.15, output: 0.6, provider: 'openai' },
     'gpt-4-turbo': { input: 10.0, output: 30.0, provider: 'openai' },
@@ -38,7 +54,18 @@ export const MODEL_PRICING = {
     'mixtral': { input: 0, output: 0, provider: 'ollama' },
     'phi3': { input: 0, output: 0, provider: 'ollama' },
     'qwen2': { input: 0, output: 0, provider: 'ollama' },
-    // OpenRouter models (provider prefixed)
+    // OpenRouter models (provider prefixed) — current generation
+    'anthropic/claude-sonnet-5': { input: 2.0, output: 10.0, provider: 'openrouter' },
+    'anthropic/claude-opus-5': { input: 5.0, output: 25.0, provider: 'openrouter' },
+    'anthropic/claude-opus-5.5': { input: 4.0, output: 20.0, provider: 'openrouter' },
+    'anthropic/claude-haiku-4.5': { input: 1.0, output: 5.0, provider: 'openrouter' },
+    'openai/gpt-6-sol': { input: 2.0, output: 10.0, provider: 'openrouter' },
+    'openai/gpt-6-luna': { input: 0.1, output: 0.5, provider: 'openrouter' },
+    'google/gemini-3.8-flash': { input: 0.75, output: 3.75, provider: 'openrouter' },
+    'meta-llama/llama-4-maverick': { input: 0.1875, output: 0.6525, provider: 'openrouter' },
+    'openai/gpt-oss-120b': { input: 0.15, output: 0.6, provider: 'openrouter' },
+    'qwen/qwen3-coder': { input: 0.3, output: 1.0, provider: 'openrouter' },
+    // OpenRouter legacy slugs (retired upstream for Claude 3.x)
     'anthropic/claude-3.5-sonnet': { input: 3.0, output: 15.0, provider: 'openrouter' },
     'anthropic/claude-3.5-sonnet:beta': { input: 3.0, output: 15.0, provider: 'openrouter' },
     'anthropic/claude-3-opus': { input: 15.0, output: 75.0, provider: 'openrouter' },
@@ -57,6 +84,8 @@ export const MODEL_PRICING = {
     'mistralai/mistral-7b-instruct': { input: 0.06, output: 0.06, provider: 'openrouter' },
     // AWS Bedrock models (ARN-style model IDs)
     // Pricing reflects Bedrock's pricing which may differ slightly from Anthropic direct
+    // Bedrock sets its own Claude pricing; this row is unverified since the
+    // 2026-09 refresh (direct API / OpenRouter Opus 4.5 is $5/$25).
     'anthropic.claude-opus-4-5-v1:0': { input: 15.0, output: 75.0, provider: 'bedrock' },
     'anthropic.claude-opus-4-v1:0': { input: 15.0, output: 75.0, provider: 'bedrock' },
     'anthropic.claude-sonnet-4-5-v2:0': { input: 3.0, output: 15.0, provider: 'bedrock' },
@@ -79,6 +108,11 @@ export const MODEL_PRICING = {
     'azure/text-embedding-3-small': { input: 0.02, output: 0, provider: 'azure-openai' },
     'azure/text-embedding-3-large': { input: 0.13, output: 0, provider: 'azure-openai' },
     // Google Gemini models (direct API)
+    'gemini-3.8-flash': { input: 0.75, output: 3.75, provider: 'gemini' },
+    'gemini-3.5-flash': { input: 1.5, output: 9.0, provider: 'gemini' },
+    'gemini-3.1-pro-preview': { input: 2.0, output: 12.0, provider: 'gemini' },
+    'gemini-2.5-flash': { input: 0.3, output: 2.5, provider: 'gemini' },
+    'gemini-2.5-pro': { input: 1.25, output: 10.0, provider: 'gemini' },
     'gemini-pro': { input: 0.5, output: 1.5, provider: 'gemini' },
     'gemini-1.0-pro': { input: 0.5, output: 1.5, provider: 'gemini' },
     'gemini-1.5-pro': { input: 3.5, output: 10.5, provider: 'gemini' },
@@ -91,6 +125,24 @@ export const MODEL_PRICING = {
     // Gemini embedding models
     'text-embedding-004': { input: 0.025, output: 0, provider: 'gemini' },
 };
+/**
+ * Resolve per-million-token pricing for a model. Exact `MODEL_PRICING` keys
+ * win; otherwise the model registry is consulted (it normalizes provider
+ * IDs such as `anthropic/claude-sonnet-5` or dated Anthropic IDs). Returns
+ * `undefined` only when neither source knows the model, which callers treat
+ * as a free/local model.
+ */
+export function resolveModelPricing(model) {
+    const pricing = MODEL_PRICING[model];
+    if (pricing) {
+        return { input: pricing.input, output: pricing.output };
+    }
+    const registryCost = getModelCost(model);
+    if (registryCost.inputCostPerMillion > 0 || registryCost.outputCostPerMillion > 0) {
+        return { input: registryCost.inputCostPerMillion, output: registryCost.outputCostPerMillion };
+    }
+    return undefined;
+}
 /**
  * Cost tracker for LLM usage
  */
@@ -106,7 +158,7 @@ export class CostTracker {
      * Calculate cost for a given usage
      */
     static calculateCost(model, usage) {
-        const pricing = MODEL_PRICING[model];
+        const pricing = resolveModelPricing(model);
         if (!pricing) {
             // Unknown model - assume Ollama (local) for safety
             return {
@@ -134,7 +186,7 @@ export class CostTracker {
      * Get cost per token for a model
      */
     static getCostPerToken(model) {
-        const pricing = MODEL_PRICING[model];
+        const pricing = resolveModelPricing(model);
         if (!pricing) {
             return { input: 0, output: 0 };
         }

package/dist/shared/llm/index.d.ts

@@ -45,13 +45,13 @@ export { type LLMProviderType, type MessageRole, type Message, type TokenUsage,
 export { ClaudeProvider, DEFAULT_CLAUDE_CONFIG, OpenAIProvider, DEFAULT_OPENAI_CONFIG, OllamaProvider, DEFAULT_OLLAMA_CONFIG, OpenRouterProvider, DEFAULT_OPENROUTER_CONFIG, GeminiProvider, DEFAULT_GEMINI_CONFIG, AzureOpenAIProvider, DEFAULT_AZURE_OPENAI_CONFIG, BedrockProvider, DEFAULT_BEDROCK_CONFIG, } from './providers';
 export { CircuitBreaker, CircuitBreakerManager, DEFAULT_CIRCUIT_BREAKER_CONFIG, } from './circuit-breaker';
 export { LLMCache, LLMResponseCache, DEFAULT_CACHE_CONFIG, type CacheableResponse, } from './cache';
-export { CostTracker, MODEL_PRICING, getGlobalCostTracker, resetGlobalCostTracker, } from './cost-tracker';
+export { CostTracker, MODEL_PRICING, resolveModelPricing, getGlobalCostTracker, resetGlobalCostTracker, } from './cost-tracker';
 export { registerProvider, tryRegisterProvider, unregisterProvider, resetProviderRegistry, isRegisteredProvider, getRegisteredProvider, listRegisteredProviders, registeredProviderTypes, createRegisteredProvider, registeredBillingDeclaration, validateProviderType, validateProviderInstance, type ProviderFactory, type ProviderRegistrationOptions, type ProviderRegistrationSource, type RegisteredProvider, } from './provider-registry';
 export { ExternalCliProvider, validateExternalCliConfig, resolvesOnPath, DEFAULT_EXTERNAL_CLI_TIMEOUT_MS, DEFAULT_EXTERNAL_CLI_CONCURRENCY, EXTERNAL_CLI_DEFAULT_MODEL, type ExternalCliProviderConfig, } from './providers/external-cli';
 export { registerExternalProviders, sanitizeExternalProviders, type ExternalProviderRegistrationResult, } from './router/external-provider-config';
 export { ProviderManager, createProviderManager, createQEProviderManager, DEFAULT_PROVIDER_MANAGER_CONFIG, } from './provider-manager';
 export { mapModelId, normalizeModelId, getCanonicalName, getModelFamily, getModelTier, isModelAvailableOnProvider, getSupportedProviders, listCanonicalModels, listModelsByProvider, listModelsByFamily, listModelsByTier, getModelMapping, MODEL_MAPPINGS, type ProviderType, type ModelMapping, } from './model-mapping';
-export { getModelCapabilities, getModelCost, getModelInfo, listModels, findModelsByCapability, findModelsByCost, findCheapestModel, findBestModelInBudget, getRecommendedModels, getDeprecationStatus, estimateRequestCost, compareModels, MODEL_REGISTRY, type ModelCapabilities, type ModelCostInfo, type ModelInfo, } from './model-registry';
+export { getModelCapabilities, getModelCost, getModelInfo, listModels, findModelsByCapability, findModelsByCost, findCheapestModel, findBestModelInBudget, getRecommendedModels, getDeprecationStatus, estimateRequestCost, compareModels, MODEL_REGISTRY, DEFAULT_SONNET_MODEL, DEFAULT_OPUS_MODEL, DEFAULT_HAIKU_MODEL, CLAUDE_TIER_MODELS, getClaudeModelForTier, RETIRING_MODELS, type ModelCapabilities, type ModelCostInfo, type ModelInfo, } from './model-registry';
 export { RouterMetricsCollector, createRouterMetricsCollector, getGlobalRouterMetrics, resetGlobalRouterMetrics, CostMetricsCollector, createCostMetricsCollector, getGlobalCostMetrics, resetGlobalCostMetrics, type RoutingDecisionRecord, type ProviderCallRecord, type FallbackRecord, type RouterMetricsSummary, type ProviderMetrics as RouterProviderMetrics, type AgentMetrics, type MetricsTimeWindow, type CostRecord, type CostTrend, type CostBreakdown, type BudgetAlert, type CostOptimizationSuggestion, } from './metrics';
 export { HybridRouter, createHybridRouter, createQERouter, } from './router';
 export type { ExtendedProviderType, RoutingMode, RoutingDecision, RouterConfig, ChatParams, ChatResponse, FallbackChain, } from './router';

package/dist/shared/llm/index.js

@@ -52,7 +52,7 @@ export { CircuitBreaker, CircuitBreakerManager, DEFAULT_CIRCUIT_BREAKER_CONFIG,
 // Cache
 export { LLMCache, LLMResponseCache, DEFAULT_CACHE_CONFIG, } from './cache.js';
 // Cost tracker
-export { CostTracker, MODEL_PRICING, getGlobalCostTracker, resetGlobalCostTracker, } from './cost-tracker.js';
+export { CostTracker, MODEL_PRICING, resolveModelPricing, getGlobalCostTracker, resetGlobalCostTracker, } from './cost-tracker.js';
 // External provider registration (ADR-127 / issue #628)
 export { registerProvider, tryRegisterProvider, unregisterProvider, resetProviderRegistry, isRegisteredProvider, getRegisteredProvider, listRegisteredProviders, registeredProviderTypes, createRegisteredProvider, registeredBillingDeclaration, validateProviderType, validateProviderInstance, } from './provider-registry.js';
 export { ExternalCliProvider, validateExternalCliConfig, resolvesOnPath, DEFAULT_EXTERNAL_CLI_TIMEOUT_MS, DEFAULT_EXTERNAL_CLI_CONCURRENCY, EXTERNAL_CLI_DEFAULT_MODEL, } from './providers/external-cli.js';
@@ -62,7 +62,7 @@ export { ProviderManager, createProviderManager, createQEProviderManager, DEFAUL
 // Model mapping (ADR-043)
 export { mapModelId, normalizeModelId, getCanonicalName, getModelFamily, getModelTier, isModelAvailableOnProvider, getSupportedProviders, listCanonicalModels, listModelsByProvider, listModelsByFamily, listModelsByTier, getModelMapping, MODEL_MAPPINGS, } from './model-mapping.js';
 // Model registry (ADR-043)
-export { getModelCapabilities, getModelCost, getModelInfo, listModels, findModelsByCapability, findModelsByCost, findCheapestModel, findBestModelInBudget, getRecommendedModels, getDeprecationStatus, estimateRequestCost, compareModels, MODEL_REGISTRY, } from './model-registry.js';
+export { getModelCapabilities, getModelCost, getModelInfo, listModels, findModelsByCapability, findModelsByCost, findCheapestModel, findBestModelInBudget, getRecommendedModels, getDeprecationStatus, estimateRequestCost, compareModels, MODEL_REGISTRY, DEFAULT_SONNET_MODEL, DEFAULT_OPUS_MODEL, DEFAULT_HAIKU_MODEL, CLAUDE_TIER_MODELS, getClaudeModelForTier, RETIRING_MODELS, } from './model-registry.js';
 // Metrics (ADR-043 Milestone 11)
 export { RouterMetricsCollector, createRouterMetricsCollector, getGlobalRouterMetrics, resetGlobalRouterMetrics, CostMetricsCollector, createCostMetricsCollector, getGlobalCostMetrics, resetGlobalCostMetrics, } from './metrics/index.js';
 // HybridRouter (ADR-043: Intelligent provider routing)

package/dist/shared/llm/interfaces.d.ts

@@ -178,7 +178,7 @@ export interface LLMConfig {
  */
 export interface ClaudeConfig extends LLMConfig {
     /** Claude model version */
-    model: 'claude-opus-4-7' | 'claude-sonnet-4-6' | 'claude-haiku-4-5-20251001' | string;
+    model: 'claude-opus-5' | 'claude-sonnet-5' | 'claude-opus-4-7' | 'claude-sonnet-4-6' | 'claude-haiku-4-5-20251001' | string;
     /** Anthropic API version */
     anthropicVersion?: string;
     /** Maximum thinking tokens for extended thinking */

package/dist/shared/llm/metrics/cost-metrics.js

@@ -9,7 +9,8 @@
  * - Cost optimization insights
  */
 import { ALL_PROVIDER_TYPES } from '../router/types.js';
-import { MODEL_PRICING } from '../cost-tracker.js';
+import { resolveModelPricing } from '../cost-tracker.js';
+import { DEFAULT_SONNET_MODEL } from '../model-registry.js';
 // ============================================================================
 // CostMetricsCollector Implementation
 // ============================================================================
@@ -244,7 +245,7 @@ export class CostMetricsCollector {
         const byModel = this.getCostByModel('24h');
         // Check for expensive models that could use cheaper alternatives
         for (const [model, breakdown] of byModel) {
-            const pricing = MODEL_PRICING[model];
+            const pricing = resolveModelPricing(model);
             if (pricing && breakdown.totalCost > 1) {
                 // Suggest cheaper alternatives
                 if (model.includes('opus')) {
@@ -253,7 +254,7 @@ export class CostMetricsCollector {
                         description: `Consider using Sonnet instead of Opus for non-critical tasks`,
                         estimatedSavings: breakdown.totalCost * 0.5,
                         currentCost: breakdown.totalCost,
-                        suggestedModel: 'claude-sonnet-4-6',
+                        suggestedModel: DEFAULT_SONNET_MODEL,
                     });
                 }
                 if (model === 'gpt-4o') {
@@ -330,7 +331,7 @@ export class CostMetricsCollector {
     // Private Helper Methods
     // ==========================================================================
     calculateCost(model, inputTokens, outputTokens) {
-        const pricing = MODEL_PRICING[model];
+        const pricing = resolveModelPricing(model);
         if (!pricing) {
             return {
                 inputCost: 0,

package/dist/shared/llm/model-mapping.js

@@ -114,6 +114,77 @@ export const MODEL_MAPPINGS = {
             bedrock: 'anthropic.claude-haiku-4-5-v1:0',
         },
     },
+    // Claude 5 generation (verified against GET /v1/models, 2026-09-27).
+    // Bedrock IDs use the Mantle `anthropic.<id>` form.
+    'claude-opus-4-6': {
+        canonical: 'Claude Opus 4.6',
+        family: 'claude',
+        tier: 'flagship',
+        providers: {
+            anthropic: 'claude-opus-4-6',
+            openrouter: 'anthropic/claude-opus-4.6',
+        },
+    },
+    'claude-opus-4-8': {
+        canonical: 'Claude Opus 4.8',
+        family: 'claude',
+        tier: 'flagship',
+        providers: {
+            anthropic: 'claude-opus-4-8',
+            openrouter: 'anthropic/claude-opus-4.8',
+            bedrock: 'anthropic.claude-opus-4-8',
+        },
+    },
+    'claude-sonnet-5': {
+        canonical: 'Claude Sonnet 5',
+        family: 'claude',
+        tier: 'standard',
+        providers: {
+            anthropic: 'claude-sonnet-5',
+            openrouter: 'anthropic/claude-sonnet-5',
+            bedrock: 'anthropic.claude-sonnet-5',
+        },
+    },
+    'claude-opus-5': {
+        canonical: 'Claude Opus 5',
+        family: 'claude',
+        tier: 'flagship',
+        providers: {
+            anthropic: 'claude-opus-5',
+            openrouter: 'anthropic/claude-opus-5',
+            bedrock: 'anthropic.claude-opus-5',
+        },
+    },
+    'claude-opus-5-5': {
+        canonical: 'Claude Opus 5.5',
+        family: 'claude',
+        tier: 'flagship',
+        providers: {
+            anthropic: 'claude-opus-5-5',
+            openrouter: 'anthropic/claude-opus-5.5',
+            bedrock: 'anthropic.claude-opus-5-5',
+        },
+    },
+    'claude-fable-5': {
+        canonical: 'Claude Fable 5',
+        family: 'claude',
+        tier: 'flagship',
+        providers: {
+            anthropic: 'claude-fable-5',
+            openrouter: 'anthropic/claude-fable-5',
+            bedrock: 'anthropic.claude-fable-5',
+        },
+    },
+    'claude-fable-5-1': {
+        canonical: 'Claude Fable 5.1',
+        family: 'claude',
+        tier: 'flagship',
+        providers: {
+            anthropic: 'claude-fable-5-1',
+            openrouter: 'anthropic/claude-fable-5.1',
+            bedrock: 'anthropic.claude-fable-5-1',
+        },
+    },
     // ==========================================================================
     // OpenAI Models
     // ==========================================================================
@@ -167,6 +238,37 @@ export const MODEL_MAPPINGS = {
             openrouter: 'openai/o1',
         },
     },
+    // GPT-6 generation (verified against OpenAI GET /v1/models, 2026-09-27)
+    'gpt-6-astra': {
+        canonical: 'GPT-6 Astra',
+        family: 'gpt',
+        tier: 'flagship',
+        providers: {
+            openai: 'gpt-6-astra',
+            azure: 'gpt-6-astra',
+            openrouter: 'openai/gpt-6-astra',
+        },
+    },
+    'gpt-6-sol': {
+        canonical: 'GPT-6 Sol',
+        family: 'gpt',
+        tier: 'standard',
+        providers: {
+            openai: 'gpt-6-sol',
+            azure: 'gpt-6-sol',
+            openrouter: 'openai/gpt-6-sol',
+        },
+    },
+    'gpt-6-luna': {
+        canonical: 'GPT-6 Luna',
+        family: 'gpt',
+        tier: 'economy',
+        providers: {
+            openai: 'gpt-6-luna',
+            azure: 'gpt-6-luna',
+            openrouter: 'openai/gpt-6-luna',
+        },
+    },
     'o1-mini': {
         canonical: 'OpenAI o1 Mini',
         family: 'gpt',
@@ -207,6 +309,35 @@ export const MODEL_MAPPINGS = {
             openrouter: 'google/gemini-flash-1.5',
         },
     },
+    // Gemini 3.x (OpenRouter catalog, 2026-09-27; Gemini API IDs assumed to
+    // match the OpenRouter slug suffix)
+    'gemini-3.8-flash': {
+        canonical: 'Gemini 3.8 Flash',
+        family: 'gemini',
+        tier: 'economy',
+        providers: {
+            gemini: 'gemini-3.8-flash',
+            openrouter: 'google/gemini-3.8-flash',
+        },
+    },
+    'gemini-3.5-flash': {
+        canonical: 'Gemini 3.5 Flash',
+        family: 'gemini',
+        tier: 'standard',
+        providers: {
+            gemini: 'gemini-3.5-flash',
+            openrouter: 'google/gemini-3.5-flash',
+        },
+    },
+    'gemini-3.1-pro-preview': {
+        canonical: 'Gemini 3.1 Pro (preview)',
+        family: 'gemini',
+        tier: 'premium',
+        providers: {
+            gemini: 'gemini-3.1-pro-preview',
+            openrouter: 'google/gemini-3.1-pro-preview',
+        },
+    },
     'gemini-ultra': {
         canonical: 'Gemini Ultra',
         family: 'gemini',

package/dist/shared/llm/model-registry.d.ts

@@ -64,6 +64,19 @@ export interface ModelCapabilities {
      * 1.0–1.35× more tokens per input at the same rate card.
      */
     tokenizerVersion?: 'legacy' | 'opus-4-7';
+    /**
+     * Whether the Messages API accepts sampling parameters (`temperature`,
+     * `top_p`, `top_k`). `false` means they are rejected with a 400
+     * ("`temperature` is deprecated for this model") — true of Opus 4.7+,
+     * Sonnet 5 and the Fable family. Omitted = accepted (legacy behavior).
+     */
+    supportsSamplingParams?: boolean;
+    /**
+     * OpenAI Chat Completions: the model rejects `max_tokens` and requires
+     * `max_completion_tokens` (GPT-5+/GPT-6 and o-series; verified live for
+     * gpt-6-sol / gpt-6-luna on 2026-09-27).
+     */
+    usesMaxCompletionTokens?: boolean;
 }
 /**
  * Cost information per million tokens
@@ -250,23 +263,36 @@ export declare function compareModels(modelA: string, modelB: string): {
     capabilityDiff: string[];
 };
 /**
- * ADR-093 Tier 3 default Sonnet for the AQE fleet.
- * Standard 200k context (not 1M). Used as the workhorse model across routing,
- * advisor fallback, and provider defaults. Replaces hardcoded 'claude-sonnet-4-20250514'
- * in 25+ call sites.
+ * Default Sonnet for the AQE fleet (routing tiers 2-3, provider defaults,
+ * advisor fallback). Claude Sonnet 5 since the 2026-09 model refresh
+ * (was claude-sonnet-4-6 under ADR-093).
  */
-export declare const DEFAULT_SONNET_MODEL = "claude-sonnet-4-6";
+export declare const DEFAULT_SONNET_MODEL = "claude-sonnet-5";
 /**
- * ADR-093 Opus escalation target. Reserved for the ADR-092 MultiModelExecutor
- * when TinyDancer sets `triggerMultiModel=true` and the task warrants maximum
- * capability. Not the Tier 3 default — that remains Sonnet 4.6 per user
- * direction 2026-04-17.
+ * Opus escalation target (routing tier 4, ADR-092 MultiModelExecutor).
+ * Claude Opus 5.5 since the 2026-09 model refresh (was claude-opus-4-7):
+ * cheaper than Opus 5 ($4/$20 vs $5/$25). Its thinking cannot be disabled,
+ * forced tool_choice (`any`/`tool`) returns 400, and its default effort is
+ * `medium`. Fable 5.1 is registered but opt-in only.
  */
-export declare const DEFAULT_OPUS_MODEL = "claude-opus-4-7";
+export declare const DEFAULT_OPUS_MODEL = "claude-opus-5-5";
 /**
- * ADR-093 economy model. Replaces retired `claude-3-5-haiku-20241022`.
+ * Economy model (routing tier 1). Replaces retired `claude-3-5-haiku-20241022`.
  */
 export declare const DEFAULT_HAIKU_MODEL = "claude-haiku-4-5";
+/**
+ * Canonical Claude model per AQE routing tier. Single source of truth for the
+ * tier maps in the router, task executor, and domain services — bump models
+ * here, not at each call site.
+ *
+ * Tier 0 is the Agent Booster (WASM) path and has no Claude model.
+ */
+export declare const CLAUDE_TIER_MODELS: Readonly<Record<1 | 2 | 3 | 4, string>>;
+/**
+ * Claude model for an AQE routing tier (1-4). Out-of-range tiers fall back to
+ * the tier 2/3 workhorse.
+ */
+export declare function getClaudeModelForTier(tier: number): string;
 /**
  * ADR-093 registry of retiring/retired model IDs with their retirement dates.
  * Used by lints and ops tooling to prevent re-introduction of deprecated IDs.