@usabledev/usable-chat 1.167.0 → 1.168.0

package/cli.js

@@ -45064,6 +45064,11 @@ var init_model_pricing = __esm({
         output: 15e-6,
         multiplier: 24
       },
+      "anthropic/claude-sonnet-5": {
+        input: 2e-6,
+        output: 1e-5,
+        multiplier: 16
+      },
       // Deepseek Models
       "deepseek/deepseek-chat": {
         input: 32e-8,
@@ -45186,6 +45191,11 @@ var init_model_pricing = __esm({
         output: 15e-7,
         multiplier: 2
       },
+      "google/gemini-3.1-flash-lite-image": {
+        input: 25e-8,
+        output: 15e-7,
+        multiplier: 2
+      },
       "google/gemini-3.1-flash-lite-preview": {
         input: 25e-8,
         output: 15e-7,
@@ -45421,8 +45431,8 @@ var init_model_pricing = __esm({
         multiplier: 1
       },
       "nvidia/nemotron-3-super-120b-a12b": {
-        input: 1e-7,
-        output: 5e-7,
+        input: 85e-9,
+        output: 4e-7,
         multiplier: 1
       },
       "nvidia/nemotron-3-ultra-550b-a55b": {
@@ -45989,6 +45999,29 @@ var init_model_registry_generated = __esm({
   "src/lib/model-registry-generated.ts"() {
     "use strict";
     GENERATED_MODELS = [
+      {
+        id: "anthropic/claude-sonnet-5",
+        displayName: "Claude Sonnet 5",
+        vendor: "anthropic",
+        category: "balanced",
+        capabilities: {
+          vision: true,
+          reasoning: false,
+          maxTokens: 128e3,
+          contextLength: 1e6
+        },
+        pricing: {
+          inputPer1M: 2,
+          outputPer1M: 10,
+          multiplier: 16
+        },
+        availability: {
+          mainChat: false,
+          experts: true,
+          embeds: true
+        },
+        createdAt: 1782843083
+      },
       {
         id: "google/gemini-3-pro-image",
         displayName: "Nano Banana Pro (Gemini 3 Pro Image)",
@@ -46691,8 +46724,8 @@ var init_model_registry_generated = __esm({
           contextLength: 1e6
         },
         pricing: {
-          inputPer1M: 0.1,
-          outputPer1M: 0.5,
+          inputPer1M: 0.085,
+          outputPer1M: 0.4,
           multiplier: 1
         },
         availability: {
@@ -47630,7 +47663,7 @@ var init_model_registry_generated = __esm({
         capabilities: {
           vision: false,
           reasoning: true,
-          maxTokens: 262144,
+          maxTokens: 100352,
           contextLength: 262144
         },
         pricing: {
@@ -48274,7 +48307,7 @@ var init_model_registry_generated = __esm({
         capabilities: {
           vision: false,
           reasoning: false,
-          maxTokens: 262144,
+          maxTokens: 100352,
           contextLength: 262144
         },
         pricing: {
@@ -48665,7 +48698,7 @@ var init_model_registry_generated = __esm({
         capabilities: {
           vision: false,
           reasoning: false,
-          maxTokens: 32768,
+          maxTokens: 100352,
           contextLength: 131072
         },
         pricing: {
@@ -49877,20 +49910,20 @@ var init_model_registry_generated = __esm({
         createdAt: 1708905600
       },
       {
-        id: "openai/gpt-4-turbo-preview",
-        displayName: "GPT-4 Turbo Preview",
+        id: "openai/gpt-3.5-turbo-0613",
+        displayName: "GPT-3.5 Turbo (older v0613)",
         vendor: "openai",
-        category: "specialized",
+        category: "fast",
         capabilities: {
           vision: false,
           reasoning: false,
           maxTokens: 4096,
-          contextLength: 128e3
+          contextLength: 4095
         },
         pricing: {
-          inputPer1M: 10,
-          outputPer1M: 30,
-          multiplier: 53
+          inputPer1M: 1,
+          outputPer1M: 2,
+          multiplier: 4
         },
         availability: {
           mainChat: false,
@@ -49900,20 +49933,20 @@ var init_model_registry_generated = __esm({
         createdAt: 1706140800
       },
       {
-        id: "openai/gpt-3.5-turbo-0613",
-        displayName: "GPT-3.5 Turbo (older v0613)",
+        id: "openai/gpt-4-turbo-preview",
+        displayName: "GPT-4 Turbo Preview",
         vendor: "openai",
-        category: "fast",
+        category: "specialized",
         capabilities: {
           vision: false,
           reasoning: false,
           maxTokens: 4096,
-          contextLength: 4095
+          contextLength: 128e3
         },
         pricing: {
-          inputPer1M: 1,
-          outputPer1M: 2,
-          multiplier: 4
+          inputPer1M: 10,
+          outputPer1M: 30,
+          multiplier: 53
         },
         availability: {
           mainChat: false,
@@ -49946,20 +49979,20 @@ var init_model_registry_generated = __esm({
         createdAt: 1693180800
       },
       {
-        id: "openai/gpt-4",
-        displayName: "GPT-4",
+        id: "openai/gpt-3.5-turbo",
+        displayName: "GPT-3.5 Turbo",
         vendor: "openai",
-        category: "specialized",
+        category: "fast",
         capabilities: {
           vision: false,
           reasoning: false,
           maxTokens: 4096,
-          contextLength: 8191
+          contextLength: 16385
         },
         pricing: {
-          inputPer1M: 30,
-          outputPer1M: 60,
-          multiplier: 120
+          inputPer1M: 0.5,
+          outputPer1M: 1.5,
+          multiplier: 3
         },
         availability: {
           mainChat: false,
@@ -49969,20 +50002,20 @@ var init_model_registry_generated = __esm({
         createdAt: 1685232e3
       },
       {
-        id: "openai/gpt-3.5-turbo",
-        displayName: "GPT-3.5 Turbo",
+        id: "openai/gpt-4",
+        displayName: "GPT-4",
         vendor: "openai",
-        category: "fast",
+        category: "specialized",
         capabilities: {
           vision: false,
           reasoning: false,
           maxTokens: 4096,
-          contextLength: 16385
+          contextLength: 8191
         },
         pricing: {
-          inputPer1M: 0.5,
-          outputPer1M: 1.5,
-          multiplier: 3
+          inputPer1M: 30,
+          outputPer1M: 60,
+          multiplier: 120
         },
         availability: {
           mainChat: false,
@@ -50074,6 +50107,7 @@ var init_model_registry = __esm({
         crossRegion: true
       },
       "claude-sonnet-4-6": { bedrockId: "anthropic.claude-sonnet-4-6", crossRegion: true },
+      "claude-sonnet-5": { bedrockId: "global.anthropic.claude-sonnet-5", crossRegion: false },
       // Opus 4.7 uses the Global Cross-Region Inference profile (no regional Geo access required)
       "claude-opus-4-7": { bedrockId: "global.anthropic.claude-opus-4-7", crossRegion: false },
       "claude-opus-4-20250514": {
@@ -146958,6 +146992,19 @@ var init_bedrock_sdk = __esm({
 });
 
 // src/lib/claude-direct.ts
+function usesAdaptiveThinking(modelId) {
+  return modelId.includes("claude-opus-4-7") || modelId.includes("claude-sonnet-5");
+}
+function supportsSamplingParameters(modelId) {
+  return !modelId.includes("claude-sonnet-5");
+}
+function prepareClaudeRequestForModel(request) {
+  if (supportsSamplingParameters(request.model)) {
+    return request;
+  }
+  const { temperature: _temperature, ...rest } = request;
+  return rest;
+}
 function convertToClaudeMessages(messages4, systemMessage) {
   const claudeMessages = [];
   let system = systemMessage;
@@ -147328,38 +147375,39 @@ function createAnthropicClient(apiKey) {
   });
 }
 async function callClaude(apiKey, request, options2) {
+  const normalizedRequest = prepareClaudeRequestForModel(request);
   const anthropic = createAnthropicClient(apiKey);
   logger.debug("claude", "Calling Claude API directly", {
-    model: request.model,
-    hasThinking: !!request.thinking,
-    thinking: request.thinking,
-    hasTools: !!request.tools,
-    toolCount: request.tools?.length || 0,
-    hasSystem: !!request.system
+    model: normalizedRequest.model,
+    hasThinking: !!normalizedRequest.thinking,
+    thinking: normalizedRequest.thinking,
+    hasTools: !!normalizedRequest.tools,
+    toolCount: normalizedRequest.tools?.length || 0,
+    hasSystem: !!normalizedRequest.system
   });
   try {
     const streamParams = {
-      model: request.model,
-      max_tokens: request.max_tokens,
-      messages: request.messages
+      model: normalizedRequest.model,
+      max_tokens: normalizedRequest.max_tokens,
+      messages: normalizedRequest.messages
     };
-    if (request.temperature !== void 0) {
-      streamParams.temperature = request.temperature;
+    if (normalizedRequest.temperature !== void 0) {
+      streamParams.temperature = normalizedRequest.temperature;
     }
-    if (request.system) {
-      streamParams.system = request.system;
+    if (normalizedRequest.system) {
+      streamParams.system = normalizedRequest.system;
     }
-    if (request.tools && request.tools.length > 0) {
-      streamParams.tools = request.tools;
-      if (request.tool_choice !== void 0 && request.tool_choice !== null) {
-        streamParams.tool_choice = request.tool_choice;
+    if (normalizedRequest.tools && normalizedRequest.tools.length > 0) {
+      streamParams.tools = normalizedRequest.tools;
+      if (normalizedRequest.tool_choice !== void 0 && normalizedRequest.tool_choice !== null) {
+        streamParams.tool_choice = normalizedRequest.tool_choice;
       }
     }
-    if (request.thinking) {
-      streamParams.thinking = request.thinking;
+    if (normalizedRequest.thinking) {
+      streamParams.thinking = normalizedRequest.thinking;
     }
-    if (request.output_config) {
-      streamParams.output_config = request.output_config;
+    if (normalizedRequest.output_config) {
+      streamParams.output_config = normalizedRequest.output_config;
     }
     const requestOptions = {
       signal: options2?.signal
@@ -147403,6 +147451,8 @@ async function callClaude(apiKey, request, options2) {
 }
 function extractClaudeModel(modelId) {
   const modelMap = {
+    // Claude Sonnet 5 models
+    "anthropic/claude-sonnet-5": "claude-sonnet-5",
     // Claude Sonnet 4.5 models
     "anthropic/claude-sonnet-4.5": "claude-sonnet-4-5-20250929",
     "anthropic/claude-sonnet-4.6": "claude-sonnet-4-6",
@@ -147457,9 +147507,9 @@ function getModelById2(modelId) {
 }
 function mapModeToModel(modeId) {
   const modeMap = {
-    "standard": "anthropic/claude-haiku-4.5",
+    standard: "anthropic/claude-haiku-4.5",
     "quick-thinking": "anthropic/claude-haiku-4.5",
-    "deep-thinking": "anthropic/claude-sonnet-4.6",
+    "deep-thinking": "anthropic/claude-sonnet-5",
     "multi-lingual": "google/gemini-3.1-pro-preview",
     "standard-fallback": "openai/gpt-5-mini"
   };
@@ -147481,6 +147531,7 @@ var init_model_config = __esm({
       "anthropic/claude-sonnet-4": { vision: true, reasoning: false },
       "anthropic/claude-sonnet-4.5": { vision: true, reasoning: true },
       "anthropic/claude-sonnet-4.6": { vision: true, reasoning: true },
+      "anthropic/claude-sonnet-5": { vision: true, reasoning: true },
       "anthropic/claude-haiku-4.5": { vision: true, reasoning: true },
       "anthropic/claude-4.5-haiku": { vision: true, reasoning: true },
       "anthropic/claude-4.5-sonnet": { vision: true, reasoning: true },
@@ -147533,9 +147584,9 @@ var init_model_tiers = __esm({
         name: "Heavy Reasoning",
         models: [
           {
-            id: "anthropic/claude-sonnet-4.6",
-            displayName: "Claude Sonnet 4.6",
-            description: "Best for professional tasks, finance, and long conversations",
+            id: "anthropic/claude-sonnet-5",
+            displayName: "Claude Sonnet 5",
+            description: "Best for professional tasks, coding, agents, and long conversations",
             provider: "Anthropic",
             costTier: "Expensive",
             zdr: true
@@ -213854,7 +213905,7 @@ HOW TO USE THIS:
 - \`model\`: **default to \`quick-thinking\` (alias for \`anthropic/claude-haiku-4.5\`) unless you have a specific reason not to.** Subagents are billed in addition to your own turn \u2014 every Sonnet-class subagent doubles cost. Guidance:
     - **\`quick-thinking\` / \`anthropic/claude-haiku-4.5\`** (cheapest, fast, strong tool use) \u2014 use this for: web/Usable search, summarisation, classification, simple extraction, drafting, tabular work, anything where one-shot quality is "good enough". Should be your **default**.
     - **\`google/gemini-3-flash-preview\`** (cheap, huge context, multimodal/multi-lingual) \u2014 use this when: you need to read a 100k+-token document, the task is multi-lingual, or there's image/audio input.
-    - **\`deep-thinking\` / \`anthropic/claude-sonnet-4.6\`** (5\u201310\xD7 the cost of Haiku) \u2014 only use when: the work requires multi-step reasoning over conflicting evidence, code generation against a specific architecture, or careful policy/risk judgement. If you find yourself reaching for this for "research", reach for Haiku first.
+    - **\`deep-thinking\` / \`anthropic/claude-sonnet-5\`** (5\u201310\xD7 the cost of Haiku) \u2014 only use when: the work requires multi-step reasoning over conflicting evidence, code generation against a specific architecture, or careful policy/risk judgement. If you find yourself reaching for this for "research", reach for Haiku first.
     - **\`google/gemini-3.1-pro-preview\`** \u2014 only when both massive context AND deep reasoning are required.
 - \`allowedTools\`: tool names exactly as they appear in YOUR own tool list. Raw MCP names (no \`mcp__usable__\` prefix) for built-in / MCP tools; \`parent_*\` prefixed names for embed-host tools registered by the surrounding application (e.g. \`parent_search_vessel_registry\`). Subagents in embed contexts CAN use parent tools \u2014 list them here just like any other tool.
 - \`systemPrompt\` (optional): override the default worker-agent system prompt.
@@ -286796,29 +286847,32 @@ async function callLLM({
     const mergedSystem = systemMessage ? claudeMessages.system ? `${claudeMessages.system}
 
 ${systemMessage}` : systemMessage : claudeMessages.system;
-    const thinkingEnabled = reasoningConfig.enabled && reasoningConfig.providerOptions?.reasoning?.max_tokens && reasoningEffort === "high";
+    const reasoningOptions = reasoningConfig.providerOptions?.reasoning;
+    const adaptiveThinkingEnabled = reasoningConfig.enabled && Boolean(reasoningOptions) && reasoningEffort === "high" && usesAdaptiveThinking(claudeModel);
+    const legacyThinkingEnabled = reasoningConfig.enabled && Boolean(reasoningOptions?.max_tokens) && reasoningEffort === "high" && !usesAdaptiveThinking(claudeModel);
+    const thinkingEnabled = adaptiveThinkingEnabled || legacyThinkingEnabled;
     const claudeRequest = {
       model: claudeModel,
       messages: claudeMessages.messages,
       system: mergedSystem,
-      max_tokens: maxTokens,
-      // Temperature must be 1 when thinking is enabled, otherwise use provided temperature
-      temperature: thinkingEnabled ? 1 : temperature
+      max_tokens: maxTokens
     };
+    if (supportsSamplingParameters(claudeModel)) {
+      claudeRequest.temperature = legacyThinkingEnabled ? 1 : temperature;
+    }
     if (shouldEnableTools && tools && tools.length > 0) {
       claudeRequest.tools = tools;
     }
-    if (thinkingEnabled && reasoningConfig.providerOptions?.reasoning?.max_tokens) {
-      const usesAdaptiveThinking = claudeModel.includes("claude-opus-4-7");
-      if (usesAdaptiveThinking) {
+    if (thinkingEnabled) {
+      if (adaptiveThinkingEnabled) {
         claudeRequest.thinking = { type: "adaptive" };
         claudeRequest.output_config = {
-          effort: reasoningEffort ?? "high"
+          effort: reasoningOptions?.effort ?? reasoningEffort ?? "high"
         };
-      } else {
+      } else if (legacyThinkingEnabled && reasoningOptions?.max_tokens) {
         claudeRequest.thinking = {
           type: "enabled",
-          budget_tokens: reasoningConfig.providerOptions.reasoning.max_tokens
+          budget_tokens: reasoningOptions.max_tokens
         };
       }
     }
@@ -308986,7 +309040,7 @@ init_tui_select();
 init_model_registry();
 
 // package.json
-var version2 = "1.167.0";
+var version2 = "1.168.0";
 
 // src/adapters/cli/model-catalog.ts
 init_codex_auth();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@usabledev/usable-chat",
-  "version": "1.167.0",
+  "version": "1.168.0",
   "description": "usable-chat — terminal harness for usable-chat (headless + TUI)",
   "type": "module",
   "bin": {