agentic-qe 3.14.3 → 3.14.4

package/dist/shared/llm/providers/azure-openai.js

@@ -29,6 +29,7 @@
  */
 import { createLLMError, } from '../interfaces.js';
 import { CostTracker } from '../cost-tracker.js';
+import { applyOpenAIParams } from '../openai-params.js';
 import { TokenMetricsCollector } from '../../../learning/token-tracker.js';
 import { toError } from '../../error-utils.js';
 import { backoffDelay } from '../retry.js';
@@ -36,7 +37,11 @@ import { backoffDelay } from '../retry.js';
  * Default Azure OpenAI configuration
  */
 export const DEFAULT_AZURE_OPENAI_CONFIG = {
-    model: 'gpt-4o', // Base model - actual deployment may vary
+    // Logical base model for cost tracking only — the deployment decides which
+    // model runs. Kept at gpt-4o so existing deployments keep their request
+    // shape; GPT-5/6 deployments must set `model` explicitly (e.g. 'gpt-6-sol')
+    // so the request uses max_completion_tokens.
+    model: 'gpt-4o',
     maxTokens: 4096,
     temperature: 0.7,
     timeoutMs: 60000,
@@ -54,8 +59,22 @@ export class AzureOpenAIProvider {
     name = 'Azure OpenAI';
     config;
     requestId = 0;
+    /** Model the caller configured explicitly (not the default), for parameter policy. */
+    explicitModel;
     constructor(config) {
         this.config = { ...DEFAULT_AZURE_OPENAI_CONFIG, ...config };
+        this.explicitModel = config.model;
+    }
+    /**
+     * Model ID used to pick the request-parameter policy. The deployment decides
+     * which model actually runs, so the default `model` is not trusted here: an
+     * explicit model wins, otherwise the deployment name is looked up. A
+     * deployment name that hides the base model (e.g. 'prod-chat') falls back to
+     * the name heuristic in openai-params, i.e. the legacy max_tokens shape —
+     * GPT-5/6 deployments with opaque names must therefore set `model`.
+     */
+    paramPolicyModel(requestModel) {
+        return requestModel ?? this.explicitModel ?? this.config.deploymentId;
     }
     /**
      * Check if Azure OpenAI is available and configured
@@ -104,10 +123,7 @@ export class AzureOpenAIProvider {
             const response = await this.fetchWithTimeout(this.buildUrl('chat/completions'), {
                 method: 'POST',
                 headers: this.getHeaders(),
-                body: JSON.stringify({
-                    max_tokens: 1,
-                    messages: [{ role: 'user', content: 'Hi' }],
-                }),
+                body: JSON.stringify(applyOpenAIParams({ messages: [{ role: 'user', content: 'Hi' }] }, this.paramPolicyModel(), { maxTokens: 1 })),
             }, 5000);
             const latencyMs = Date.now() - start;
             if (!response.ok) {
@@ -152,11 +168,7 @@ export class AzureOpenAIProvider {
         const temperature = options?.temperature ?? this.config.temperature ?? 0.7;
         const requestId = `azure-openai-${++this.requestId}-${Date.now()}`;
         const start = Date.now();
-        const body = {
-            max_tokens: maxTokens,
-            temperature,
-            messages,
-        };
+        const body = applyOpenAIParams({ messages }, this.paramPolicyModel(options?.model), { maxTokens, temperature });
         if (options?.stopSequences && options.stopSequences.length > 0) {
             body.stop = options.stopSequences;
         }
@@ -291,11 +303,10 @@ export class AzureOpenAIProvider {
      */
     getSupportedModels() {
         return [
-            'gpt-4o',
-            'gpt-4o-mini',
-            'gpt-4-turbo',
-            'gpt-4',
-            'gpt-35-turbo', // Azure uses different naming
+            'gpt-6-sol',
+            'gpt-6-luna',
+            'gpt-6-astra',
+            'gpt-5.4-mini',
             'text-embedding-ada-002',
             'text-embedding-3-small',
             'text-embedding-3-large',

package/dist/shared/llm/providers/bedrock.js

@@ -16,11 +16,14 @@ import { CostTracker } from '../cost-tracker.js';
 import { TokenMetricsCollector } from '../../../learning/token-tracker.js';
 import { toError } from '../../error-utils.js';
 import { backoffDelay } from '../retry.js';
+import { applyAnthropicParams } from '../anthropic-params.js';
 /**
  * Default Bedrock configuration
  */
 export const DEFAULT_BEDROCK_CONFIG = {
-    model: 'anthropic.claude-3-5-sonnet-20241022-v2:0',
+    // Sonnet 4.6 is the newest model verified on the InvokeModel path this
+    // provider uses; Claude 5 models are documented on Bedrock's Mantle endpoint.
+    model: 'anthropic.claude-sonnet-4-6-v1:0',
     region: 'us-east-1',
     maxTokens: 4096,
     temperature: 0.7,
@@ -41,24 +44,12 @@ export const BEDROCK_MODEL_MAPPING = {
     'claude-sonnet-4-6': 'anthropic.claude-sonnet-4-6-v1:0',
     'claude-haiku-4-5-20251001': 'anthropic.claude-haiku-4-5-v1:0',
     'claude-haiku-4-5': 'anthropic.claude-haiku-4-5-v1:0',
-    // Retiring 2026-06-15 — kept for existing traffic during transition
+    // Previous generation, still served by Anthropic
     'claude-opus-4-5-20251101': 'anthropic.claude-opus-4-5-v1:0',
     'claude-opus-4-5': 'anthropic.claude-opus-4-5-v1:0',
-    'claude-opus-4-20250514': 'anthropic.claude-opus-4-v1:0',
-    'claude-opus-4': 'anthropic.claude-opus-4-v1:0',
-    'claude-sonnet-4-20250514': 'anthropic.claude-sonnet-4-v1:0',
-    'claude-sonnet-4': 'anthropic.claude-sonnet-4-v1:0',
-    'claude-3-5-haiku-20241022': 'anthropic.claude-3-5-haiku-v1:0',
-    'claude-haiku-3-5': 'anthropic.claude-3-5-haiku-v1:0',
-    // Claude Sonnet 4.5
     'claude-sonnet-4-5-20250929': 'anthropic.claude-sonnet-4-5-v2:0',
     'claude-sonnet-4-5': 'anthropic.claude-sonnet-4-5-v2:0',
-    // Legacy Claude 3 models
-    'claude-3-opus-20240229': 'anthropic.claude-3-opus-20240229-v1:0',
-    'claude-3-sonnet-20240229': 'anthropic.claude-3-sonnet-20240229-v1:0',
-    'claude-3-haiku-20240307': 'anthropic.claude-3-haiku-20240307-v1:0',
-    // Claude 3.5 Sonnet
-    'claude-3-5-sonnet-20241022': 'anthropic.claude-3-5-sonnet-20241022-v2:0',
+    // Claude 3.x, Sonnet 4 and Opus 4/4.1 are retired and intentionally absent.
 };
 /**
  * Reverse mapping from Bedrock model IDs to canonical IDs
@@ -160,10 +151,11 @@ export class BedrockProvider {
             max_tokens: maxTokens,
             messages,
         };
-        // Temperature must be between 0 and 1 for Bedrock
-        if (temperature !== undefined) {
-            body.temperature = Math.min(Math.max(temperature, 0), 1);
-        }
+        // Temperature must be between 0 and 1 for Bedrock, and newer Claude
+        // models reject it entirely.
+        applyAnthropicParams(body, bedrockModelId, {
+            temperature: temperature !== undefined ? Math.min(Math.max(temperature, 0), 1) : undefined,
+        });
         if (options?.systemPrompt) {
             body.system = options.systemPrompt;
         }

package/dist/shared/llm/providers/claude.js

@@ -10,13 +10,13 @@ import { CostTracker } from '../cost-tracker.js';
 import { TokenMetricsCollector } from '../../../learning/token-tracker.js';
 import { toError } from '../../error-utils.js';
 import { backoffDelay } from '../retry.js';
-import { resolveEffortLevel, downgradeEffort } from '../effort-resolver.js';
-import { getModelCapabilities } from '../model-registry.js';
+import { applyAnthropicParams } from '../anthropic-params.js';
+import { DEFAULT_SONNET_MODEL } from '../model-registry.js';
 /**
  * Default Claude configuration
  */
 export const DEFAULT_CLAUDE_CONFIG = {
-    model: 'claude-sonnet-4-6',
+    model: DEFAULT_SONNET_MODEL,
     maxTokens: 4096,
     temperature: 0.7,
     timeoutMs: 60000,
@@ -118,9 +118,11 @@ export class ClaudeProvider {
         const body = {
             model,
             max_tokens: maxTokens,
-            temperature,
             messages,
         };
+        // Sampling params and effort are model-dependent: current Claude models
+        // reject `temperature` and take effort only as `output_config.effort`.
+        applyAnthropicParams(body, model, { temperature, effort: options?.effort });
         if (options?.systemPrompt) {
             // ADR-088: when caching is enabled, send the system prompt as a
             // content-block array with an ephemeral cache breakpoint — repeated
@@ -135,31 +137,6 @@ export class ClaudeProvider {
         if (options?.stopSequences && options.stopSequences.length > 0) {
             body.stop_sequences = options.stopSequences;
         }
-        // ADR-093: apply effort level only on models that actually advertise it.
-        //
-        // Anthropic's Messages API accepts effort nested under the `thinking`
-        // block, not as a top-level field. We only set it when the target model
-        // has `supportsEffortXHigh: true` in the registry (= Opus 4.7 today).
-        // For every other model we send nothing thinking/effort-related — the
-        // caller's `options.effort` is ignored rather than silently downgraded
-        // to a value the model may not accept.
-        //
-        // Rationale: the top-level `effort` shape was unverified against
-        // Anthropic's public schema and risked 400-ing Sonnet 4.6 / Haiku 4.5
-        // calls. Gating behind the capability flag keeps behavior safe on
-        // models without xhigh support; Opus 4.7 gets the advertised quality
-        // boost once we verify the exact nested shape Anthropic accepts.
-        try {
-            const caps = getModelCapabilities(model);
-            if (caps.supportsEffortXHigh) {
-                const requested = resolveEffortLevel({ override: options?.effort });
-                const effort = downgradeEffort(requested, 'xhigh');
-                body.thinking = { type: 'adaptive', effort };
-            }
-        }
-        catch {
-            // Model not in registry — skip effort entirely.
-        }
         try {
             const response = await this.fetchWithRetry(`${this.getBaseUrl()}/v1/messages`, {
                 method: 'POST',

package/dist/shared/llm/providers/codex.js

@@ -174,9 +174,10 @@ export class CodexProvider {
         return { ...this.config };
     }
     getSupportedModels() {
-        // Codex resolves the concrete model from its profile/login; these are the
-        // common ids accepted via -m. `default` = whatever the CLI is configured to.
-        return [CODEX_DEFAULT_MODEL, 'gpt-5-codex', 'o3', 'o4-mini'];
+        // Informational only — not used to validate. Codex resolves the concrete
+        // model from its profile/login and validates any id passed via -m itself.
+        // `default` = whatever the CLI is configured to.
+        return [CODEX_DEFAULT_MODEL, 'gpt-6-sol', 'gpt-6-luna', 'gpt-6-astra', 'gpt-5.6-sol', 'gpt-5.3-codex'];
     }
     getCostPerToken() {
         return { input: 0, output: 0 };

package/dist/shared/llm/providers/gemini.d.ts

@@ -13,7 +13,7 @@ import { LLMProvider, LLMProviderType, LLMConfig, Message, GenerateOptions, Embe
  * Gemini-specific configuration
  */
 export interface GeminiConfig extends LLMConfig {
-    /** Gemini model (e.g., 'gemini-pro', 'gemini-1.5-pro', 'gemini-1.5-flash') */
+    /** Gemini model (e.g., 'gemini-2.5-flash', 'gemini-2.5-pro', 'gemini-3.8-flash') */
     model: string;
     /** Safety settings threshold */
     safetyThreshold?: 'BLOCK_NONE' | 'BLOCK_ONLY_HIGH' | 'BLOCK_MEDIUM_AND_ABOVE' | 'BLOCK_LOW_AND_ABOVE';
@@ -24,7 +24,8 @@ export interface GeminiConfig extends LLMConfig {
 export declare const DEFAULT_GEMINI_CONFIG: GeminiConfig;
 /**
  * Gemini model pricing (cost per 1M tokens in USD)
- * Prices as of early 2025
+ * 3.x prices from the OpenRouter catalog (2026-09-27); 3.x IDs other than
+ * gemini-3.5-flash are not yet verified against the Gemini ListModels API.
  */
 export declare const GEMINI_PRICING: Record<string, {
     input: number;

package/dist/shared/llm/providers/gemini.js

@@ -29,24 +29,28 @@ export const DEFAULT_GEMINI_CONFIG = {
 };
 /**
  * Gemini model pricing (cost per 1M tokens in USD)
- * Prices as of early 2025
+ * 3.x prices from the OpenRouter catalog (2026-09-27); 3.x IDs other than
+ * gemini-3.5-flash are not yet verified against the Gemini ListModels API.
  */
 export const GEMINI_PRICING = {
-    // Gemini Pro models
+    // Gemini 3.x (current generation)
+    'gemini-3.8-flash': { input: 0.75, output: 3.75 },
+    'gemini-3.5-flash': { input: 1.5, output: 9.0 },
+    'gemini-3.5-flash-lite': { input: 0.3, output: 2.5 },
+    'gemini-3.1-pro-preview': { input: 2.0, output: 12.0 },
+    // Gemini 2.5
+    'gemini-2.5-pro': { input: 1.25, output: 10.0 },
+    'gemini-2.5-flash': { input: 0.3, output: 2.5 },
+    'gemini-flash-latest': { input: 0.3, output: 2.5 },
+    // Deprecated: retired by Google. Kept only so historical usage records
+    // still price correctly — never use as a default.
     'gemini-pro': { input: 0.5, output: 1.5 },
     'gemini-1.0-pro': { input: 0.5, output: 1.5 },
-    // Gemini 1.5 Pro models
     'gemini-1.5-pro': { input: 3.5, output: 10.5 },
     'gemini-1.5-pro-latest': { input: 3.5, output: 10.5 },
-    // Gemini Flash models (economical)
     'gemini-1.5-flash': { input: 0.075, output: 0.3 },
     'gemini-1.5-flash-latest': { input: 0.075, output: 0.3 },
     'gemini-2.0-flash-exp': { input: 0.075, output: 0.3 },
-    // Gemini 2.5 (current generation)
-    'gemini-2.5-pro': { input: 1.25, output: 10.0 },
-    'gemini-2.5-flash': { input: 0.3, output: 2.5 },
-    'gemini-flash-latest': { input: 0.3, output: 2.5 },
-    // Gemini Ultra (flagship)
     'gemini-ultra': { input: 7.0, output: 21.0 },
     'gemini-1.0-ultra': { input: 7.0, output: 21.0 },
     // Default fallback
@@ -399,16 +403,21 @@ export class GeminiProvider {
         }
         // Current text-generation models (verified via Gemini ListModels API 2026-06-27).
         // Prefer the `-latest` aliases as defaults so they don't rot (gemini-1.5-pro was retired).
+        // gemini-3.8-flash, gemini-3.5-flash-lite and gemini-3.1-pro-preview come from the
+        // OpenRouter catalog (2026-09-27) and are not yet verified against ListModels.
         return [
             'gemini-flash-latest',
             'gemini-pro-latest',
             'gemini-flash-lite-latest',
+            'gemini-3.8-flash',
+            'gemini-3.5-flash',
+            'gemini-3.5-flash-lite',
+            'gemini-3.1-pro-preview',
             'gemini-2.5-pro',
             'gemini-2.5-flash',
             'gemini-2.5-flash-lite',
             'gemini-2.0-flash',
             'gemini-2.0-flash-lite',
-            'gemini-3.5-flash',
         ];
     }
     /**

package/dist/shared/llm/providers/ollama.d.ts

@@ -19,6 +19,8 @@ export declare class OllamaProvider implements LLMProvider {
     private config;
     private requestId;
     private availableModels;
+    /** Full installed names including tag (e.g. 'qwen3-coder:30b'), from the last health check. */
+    private installedTags;
     /** EMA of observed generation throughput (tok/s) per model, for adaptive timeouts (#3). */
     private observedTokPerSec;
     constructor(config?: Partial<OllamaConfig>);

package/dist/shared/llm/providers/ollama.js

@@ -14,7 +14,7 @@ import { computeAdaptiveTimeoutMs } from './ollama-timeout.js';
  * Default Ollama configuration
  */
 export const DEFAULT_OLLAMA_CONFIG = {
-    model: 'llama3.1',
+    model: 'qwen3-coder:30b', // ADR-111 local QE floor — never qwen3:8b
     baseUrl: resolveOllamaBaseUrl('http://localhost:11434'),
     maxTokens: 4096,
     temperature: 0.7,
@@ -34,6 +34,8 @@ export class OllamaProvider {
     config;
     requestId = 0;
     availableModels = [];
+    /** Full installed names including tag (e.g. 'qwen3-coder:30b'), from the last health check. */
+    installedTags = [];
     /** EMA of observed generation throughput (tok/s) per model, for adaptive timeouts (#3). */
     observedTokPerSec = new Map();
     constructor(config = {}) {
@@ -92,12 +94,17 @@ export class OllamaProvider {
             }
             const data = await response.json();
             // Extract available model names
-            this.availableModels =
-                data.models?.map((m) => m.name.split(':')[0]) ?? [];
-            // Check if the configured model is available
-            const hasConfiguredModel = this.availableModels.some((m) => m === this.config.model ||
-                m.startsWith(this.config.model) ||
-                this.config.model.startsWith(m));
+            this.installedTags = data.models?.map((m) => m.name) ?? [];
+            this.availableModels = this.installedTags.map((name) => name.split(':')[0]);
+            // Check if the configured model is available. A tagged model
+            // ('qwen3-coder:30b') must match an installed tag exactly; prefix
+            // matching reported 'qwen3:8b' as satisfying 'qwen3-coder:30b', a false
+            // healthy that only failed later at generate time. An untagged model
+            // ('llama3.1') matches any installed tag of that name.
+            const configured = this.config.model;
+            const hasConfiguredModel = configured.includes(':')
+                ? this.installedTags.includes(configured)
+                : this.availableModels.includes(configured);
             if (!hasConfiguredModel && this.availableModels.length > 0) {
                 return {
                     healthy: true,
@@ -269,7 +276,7 @@ export class OllamaProvider {
      */
     async complete(prompt, options) {
         // Use generate with code-completion-optimized settings
-        // Prefer codellama for code completion if available
+        // Prefer a code-tuned model for completion if one is installed
         const model = options?.model ?? this.selectCodeModel();
         const response = await this.generate(prompt, {
             model,
@@ -297,16 +304,15 @@ export class OllamaProvider {
      */
     getSupportedModels() {
         // Return known models plus any detected from health check
+        // Curated current families; installed models are merged in below.
+        // Tag availability varies by Ollama library version — pull to confirm.
         const knownModels = [
-            'llama3',
-            'llama3.1',
-            'llama3.2',
-            'codellama',
-            'mistral',
-            'mixtral',
-            'phi3',
-            'qwen2',
-            'gemma',
+            'qwen3-coder:30b',
+            'qwen3:30b-a3b',
+            'gpt-oss:20b',
+            'gpt-oss:120b',
+            'llama4:scout',
+            'gemma4',
             'nomic-embed-text',
         ];
         const uniqueModels = new Set([...knownModels, ...this.availableModels]);
@@ -368,10 +374,14 @@ export class OllamaProvider {
      * Select the best available code model
      */
     selectCodeModel() {
-        const codeModels = ['codellama', 'llama3.1', 'llama3', 'mistral'];
+        // Most-preferred first. Deliberately excludes qwen3:8b (below the ADR-111 floor).
+        const codeModels = ['qwen3-coder', 'gpt-oss', 'qwen3:30b-a3b', 'codellama'];
         for (const model of codeModels) {
-            if (this.availableModels.some((m) => m === model || m.startsWith(model))) {
-                return model;
+            // Return the full installed tag (e.g. 'qwen3-coder:30b'), not the prefix — a bare
+            // family name resolves to ':latest', which may not be the tag that is pulled.
+            const installed = this.installedTags.find((m) => m === model || m.startsWith(model));
+            if (installed) {
+                return installed;
             }
         }
         return this.config.model;

package/dist/shared/llm/providers/openai.js

@@ -7,6 +7,7 @@
  */
 import { createLLMError, } from '../interfaces.js';
 import { CostTracker } from '../cost-tracker.js';
+import { applyOpenAIParams } from '../openai-params.js';
 import { TokenMetricsCollector } from '../../../learning/token-tracker.js';
 import { toError } from '../../error-utils.js';
 import { backoffDelay } from '../retry.js';
@@ -14,7 +15,7 @@ import { backoffDelay } from '../retry.js';
  * Default OpenAI configuration
  */
 export const DEFAULT_OPENAI_CONFIG = {
-    model: 'gpt-4o',
+    model: 'gpt-6-sol',
     maxTokens: 4096,
     temperature: 0.7,
     timeoutMs: 60000,
@@ -66,11 +67,7 @@ export class OpenAIProvider {
             const response = await this.fetchWithTimeout(`${this.getBaseUrl()}/v1/chat/completions`, {
                 method: 'POST',
                 headers: this.getHeaders(),
-                body: JSON.stringify({
-                    model: this.config.model,
-                    max_tokens: 1,
-                    messages: [{ role: 'user', content: 'Hi' }],
-                }),
+                body: JSON.stringify(applyOpenAIParams({ model: this.config.model, messages: [{ role: 'user', content: 'Hi' }] }, this.config.model, { maxTokens: 1 })),
             }, 5000);
             const latencyMs = Date.now() - start;
             if (!response.ok) {
@@ -112,12 +109,7 @@ export class OpenAIProvider {
         const temperature = options?.temperature ?? this.config.temperature ?? 0.7;
         const requestId = `openai-${++this.requestId}-${Date.now()}`;
         const start = Date.now();
-        const body = {
-            model,
-            max_tokens: maxTokens,
-            temperature,
-            messages,
-        };
+        const body = applyOpenAIParams({ model, messages }, model, { maxTokens, temperature });
         if (options?.stopSequences && options.stopSequences.length > 0) {
             body.stop = options.stopSequences;
         }
@@ -248,11 +240,12 @@ export class OpenAIProvider {
      */
     getSupportedModels() {
         return [
-            'gpt-4o',
-            'gpt-4o-mini',
-            'gpt-4-turbo',
-            'gpt-4',
-            'gpt-3.5-turbo',
+            'gpt-6-sol',
+            'gpt-6-luna',
+            'gpt-6-astra',
+            'gpt-5.6-sol',
+            'gpt-5.6-luna',
+            'gpt-5.4-mini',
         ];
     }
     /**

package/dist/shared/llm/providers/openrouter.d.ts

@@ -14,7 +14,7 @@ import { LLMProvider, LLMProviderType, LLMConfig, Message, GenerateOptions, Embe
  * OpenRouter-specific configuration
  */
 export interface OpenRouterConfig extends LLMConfig {
-    /** OpenRouter model (e.g., 'anthropic/claude-3.5-sonnet', 'openai/gpt-4o') */
+    /** OpenRouter model (e.g., 'anthropic/claude-sonnet-5', 'openai/gpt-6-sol') */
     model: string;
     /** Site URL for OpenRouter attribution (optional but recommended) */
     siteUrl?: string;

package/dist/shared/llm/providers/openrouter.js

@@ -14,11 +14,12 @@ import { TokenMetricsCollector } from '../../../learning/token-tracker.js';
 import { toError } from '../../error-utils.js';
 import { backoffDelay } from '../retry.js';
 import { safeJsonParse } from '../../safe-json.js';
+import { resolveModelPricing } from '../cost-tracker.js';
 /**
  * Default OpenRouter configuration
  */
 export const DEFAULT_OPENROUTER_CONFIG = {
-    model: 'anthropic/claude-3.5-sonnet',
+    model: 'anthropic/claude-sonnet-5',
     maxTokens: 4096,
     temperature: 0.7,
     timeoutMs: 60000,
@@ -34,29 +35,62 @@ export const DEFAULT_OPENROUTER_CONFIG = {
  */
 export const OPENROUTER_PRICING = {
     // Anthropic models via OpenRouter
+    'anthropic/claude-sonnet-5': { input: 2.0, output: 10.0 },
+    'anthropic/claude-opus-5': { input: 5.0, output: 25.0 },
+    'anthropic/claude-opus-5.5': { input: 4.0, output: 20.0 },
+    'anthropic/claude-haiku-4.5': { input: 1.0, output: 5.0 },
+    // OpenAI models via OpenRouter
+    'openai/gpt-6-astra': { input: 10.0, output: 50.0 },
+    'openai/gpt-6-sol': { input: 2.0, output: 10.0 },
+    'openai/gpt-6-luna': { input: 0.1, output: 0.5 },
+    'openai/gpt-oss-120b': { input: 0.15, output: 0.6 },
+    // Google models via OpenRouter
+    'google/gemini-3.8-flash': { input: 0.75, output: 3.75 },
+    'google/gemini-3.5-flash': { input: 1.5, output: 9.0 },
+    'google/gemini-3.5-flash-lite': { input: 0.3, output: 2.5 },
+    'google/gemini-3.1-pro-preview': { input: 2.0, output: 12.0 },
+    'google/gemini-2.5-pro': { input: 1.25, output: 10.0 },
+    'google/gemini-2.5-flash': { input: 0.3, output: 2.5 },
+    // Meta models via OpenRouter
+    'meta-llama/llama-4-maverick': { input: 0.1875, output: 0.6525 },
+    'meta-llama/llama-4-scout': { input: 0.1, output: 0.3 },
+    // Mistral models via OpenRouter
+    'mistralai/mistral-large-2512': { input: 0.5, output: 1.5 },
+    // DeepSeek / Qwen models via OpenRouter
+    'deepseek/deepseek-v4-pro': { input: 0.348, output: 0.696 },
+    'deepseek/deepseek-v4.1-flash': { input: 0.035, output: 0.29 },
+    'qwen/qwen3-coder': { input: 0.3, output: 1.0 },
+    'openai/gpt-oss-20b': { input: 0.018, output: 0.09 },
+    // Deprecated: no longer served by OpenRouter. Kept only so historical
+    // usage records still price correctly — never use as a default.
     'anthropic/claude-3.5-sonnet': { input: 3.0, output: 15.0 },
     'anthropic/claude-3.5-sonnet:beta': { input: 3.0, output: 15.0 },
     'anthropic/claude-3-opus': { input: 15.0, output: 75.0 },
     'anthropic/claude-3-sonnet': { input: 3.0, output: 15.0 },
     'anthropic/claude-3-haiku': { input: 0.25, output: 1.25 },
-    // OpenAI models via OpenRouter
     'openai/gpt-4o': { input: 5.0, output: 15.0 },
     'openai/gpt-4o-mini': { input: 0.15, output: 0.6 },
     'openai/gpt-4-turbo': { input: 10.0, output: 30.0 },
     'openai/gpt-3.5-turbo': { input: 0.5, output: 1.5 },
-    // Google models via OpenRouter
     'google/gemini-pro': { input: 0.125, output: 0.375 },
     'google/gemini-pro-1.5': { input: 3.5, output: 10.5 },
-    // Meta models via OpenRouter
     'meta-llama/llama-3.1-70b-instruct': { input: 0.59, output: 0.79 },
     'meta-llama/llama-3.1-8b-instruct': { input: 0.06, output: 0.06 },
-    // Mistral models via OpenRouter
     'mistralai/mistral-large': { input: 2.0, output: 6.0 },
     'mistralai/mixtral-8x7b-instruct': { input: 0.24, output: 0.24 },
     'mistralai/mistral-7b-instruct': { input: 0.06, output: 0.06 },
     // Default fallback for unknown models
     'default': { input: 1.0, output: 3.0 },
 };
+/**
+ * Per-million-token pricing for an OpenRouter slug. The local table wins;
+ * otherwise the shared cost tracker / model registry (which normalize
+ * `anthropic/…`, `openai/…` slugs) are consulted before the generic default,
+ * so routable models are never billed at the placeholder rate.
+ */
+function openRouterPricing(model) {
+    return OPENROUTER_PRICING[model] ?? resolveModelPricing(model) ?? OPENROUTER_PRICING['default'];
+}
 /**
  * OpenRouter LLM provider implementation
  */
@@ -146,6 +180,7 @@ export class OpenRouterProvider {
         const temperature = options?.temperature ?? this.config.temperature ?? 0.7;
         const requestId = `openrouter-${++this.requestId}-${Date.now()}`;
         const start = Date.now();
+        // Sent as-is (no openai-params gating): OpenRouter drops params a model's catalog entry doesn't support.
         const body = {
             model,
             max_tokens: maxTokens,
@@ -399,27 +434,28 @@ export class OpenRouterProvider {
         }
         // Return common models as fallback
         return [
-            'anthropic/claude-3.5-sonnet',
-            'anthropic/claude-3-opus',
-            'anthropic/claude-3-sonnet',
-            'anthropic/claude-3-haiku',
-            'openai/gpt-4o',
-            'openai/gpt-4o-mini',
-            'openai/gpt-4-turbo',
-            'openai/gpt-3.5-turbo',
-            'google/gemini-pro',
-            'google/gemini-pro-1.5',
-            'meta-llama/llama-3.1-70b-instruct',
-            'meta-llama/llama-3.1-8b-instruct',
-            'mistralai/mistral-large',
-            'mistralai/mixtral-8x7b-instruct',
+            'anthropic/claude-sonnet-5',
+            'anthropic/claude-opus-5.5',
+            'anthropic/claude-opus-5',
+            'anthropic/claude-haiku-4.5',
+            'openai/gpt-6-sol',
+            'openai/gpt-6-luna',
+            'openai/gpt-6-astra',
+            'google/gemini-3.8-flash',
+            'google/gemini-3.1-pro-preview',
+            'google/gemini-2.5-flash',
+            'meta-llama/llama-4-maverick',
+            'deepseek/deepseek-v4-pro',
+            'qwen/qwen3-coder',
+            'mistralai/mistral-large-2512',
+            'openai/gpt-oss-120b',
         ];
     }
     /**
      * Get cost per token for current model
      */
     getCostPerToken() {
-        const pricing = OPENROUTER_PRICING[this.config.model] || OPENROUTER_PRICING['default'];
+        const pricing = openRouterPricing(this.config.model);
         return {
             input: pricing.input / 1_000_000,
             output: pricing.output / 1_000_000,
@@ -435,7 +471,7 @@ export class OpenRouterProvider {
      * Calculate cost for a given usage
      */
     calculateCost(model, usage) {
-        const pricing = OPENROUTER_PRICING[model] || OPENROUTER_PRICING['default'];
+        const pricing = openRouterPricing(model);
         // Convert from per-million to actual cost
         const inputCost = (usage.promptTokens / 1_000_000) * pricing.input;
         const outputCost = (usage.completionTokens / 1_000_000) * pricing.output;