@usabledev/usable-chat 1.215.12 → 1.216.0

package/cli.js

@@ -45618,11 +45618,19 @@ var init_tui_select = __esm({
   }
 });
 
+// src/lib/haiku-model-pricing.ts
+var init_haiku_model_pricing = __esm({
+  "src/lib/haiku-model-pricing.ts"() {
+    "use strict";
+  }
+});
+
 // src/lib/image-model-pricing.ts
 var OPENAI_IMAGE_TEXT_INPUT_PER_TOKEN, OPENAI_IMAGE_OUTPUT_PER_TOKEN, OPENAI_IMAGE_COST_MULTIPLIER, OPENAI_IMAGE_RATES, OPENAI_IMAGE_MODEL_PRICING;
 var init_image_model_pricing = __esm({
   "src/lib/image-model-pricing.ts"() {
     "use strict";
+    init_haiku_model_pricing();
     OPENAI_IMAGE_TEXT_INPUT_PER_TOKEN = 5e-6;
     OPENAI_IMAGE_OUTPUT_PER_TOKEN = 3e-5;
     OPENAI_IMAGE_COST_MULTIPLIER = 50;
@@ -45707,6 +45715,11 @@ var init_model_pricing = __esm({
         output: 5e-5,
         multiplier: 80
       },
+      "anthropic/claude-haiku-5.5": {
+        input: 1e-7,
+        output: 5e-7,
+        multiplier: 1
+      },
       "anthropic/claude-haiku-4.5": {
         input: 1e-6,
         output: 5e-6,
@@ -47009,6 +47022,16 @@ var init_model_registry_generated = __esm({
   "src/lib/model-registry-generated.ts"() {
     "use strict";
     GENERATED_MODELS = [
+      {
+        id: "anthropic/claude-haiku-5.5",
+        displayName: "Claude Haiku 5.5",
+        vendor: "anthropic",
+        category: "fast",
+        capabilities: { vision: true, reasoning: true, maxTokens: 128e3, contextLength: 1e6 },
+        pricing: { inputPer1M: 0.1, outputPer1M: 0.5, multiplier: 1 },
+        availability: { mainChat: false, experts: true, embeds: true },
+        createdAt: 1791397883
+      },
       {
         id: "openai/gpt-6.1-sol",
         displayName: "GPT-6.1 Sol",
@@ -52715,6 +52738,11 @@ var init_model_registry = __esm({
     init_model_registry_generated();
     BEDROCK_MODEL_DEFS = {
       // Claude 4.x+ — require cross-region inference profiles
+      // Launch-day Geo CRIS profile, using the documented unsuffixed model ID.
+      "claude-haiku-5-5": {
+        bedrockId: "anthropic.claude-haiku-5-5",
+        crossRegion: true
+      },
       "claude-haiku-4-5-20251001": {
         bedrockId: "anthropic.claude-haiku-4-5-20251001-v1:0",
         crossRegion: true
@@ -138231,7 +138259,7 @@ function normalizeClaudeModelId(modelId) {
 }
 function usesAdaptiveThinking(modelId) {
   const id = normalizeClaudeModelId(modelId);
-  return id.includes("claude-opus-4-7") || id.includes("claude-opus-4-8") || id.includes("claude-opus-5") || id.includes("claude-sonnet-5") || id.includes("claude-fable-5");
+  return id.includes("claude-opus-4-7") || id.includes("claude-opus-4-8") || id.includes("claude-opus-5") || id.includes("claude-sonnet-5") || id.includes("claude-fable-5") || id.includes("claude-haiku-5");
 }
 function supportsSamplingParameters(modelId) {
   return !usesAdaptiveThinking(modelId);
@@ -138243,6 +138271,17 @@ function prepareClaudeRequestForModel(request) {
   const { temperature: _temperature, ...rest } = request;
   return rest;
 }
+function getClaudeThinkingConfig(model, enabled2, effort, legacyBudget) {
+  const isHaiku5 = normalizeClaudeModelId(model).includes("claude-haiku-5");
+  if (isHaiku5 && (!enabled2 || !effort)) {
+    return { thinking: { type: "disabled" }, output_config: { effort: "medium" } };
+  }
+  if (!enabled2 || !isHaiku5 && effort !== "high") return {};
+  if (usesAdaptiveThinking(model)) {
+    return { thinking: { type: "adaptive" }, output_config: { effort: effort ?? "medium" } };
+  }
+  return legacyBudget ? { thinking: { type: "enabled", budget_tokens: legacyBudget } } : {};
+}
 function convertToClaudeMessages(messages4, systemMessage) {
   const claudeMessages = [];
   let system = systemMessage;
@@ -138738,6 +138777,7 @@ function extractClaudeModel(modelId) {
     // Claude Fable models
     "anthropic/claude-fable-5": "claude-fable-5",
     "anthropic/claude-fable-5.1": "claude-fable-5-1",
+    "anthropic/claude-haiku-5.5": "claude-haiku-5-5",
     // Claude Haiku 4.5 models
     "anthropic/claude-haiku-4.5": "claude-haiku-4-5-20251001",
     "anthropic/claude-4.5-haiku": "claude-haiku-4-5-20251001",
@@ -175712,8 +175752,9 @@ var init_model_tiers = __esm({
         name: "General Purpose",
         models: [
           {
-            id: "anthropic/claude-haiku-4.5",
-            displayName: "Claude Haiku 4.5",
+            id: "anthropic/claude-haiku-5.5",
+            effort: "medium",
+            displayName: "Claude Haiku 5.5",
             description: "Fast agentic responses, strong tool use and instruction following",
             provider: "Anthropic",
             zdr: true
@@ -176532,8 +176573,8 @@ function getModelById2(modelId) {
 }
 function mapModeToModel(modeId) {
   const modeMap = {
-    standard: "anthropic/claude-haiku-4.5",
-    "quick-thinking": "anthropic/claude-haiku-4.5",
+    standard: "anthropic/claude-haiku-5.5",
+    "quick-thinking": "anthropic/claude-haiku-5.5",
     "deep-thinking": "anthropic/claude-sonnet-5.5",
     "multi-lingual": "google/gemini-3.1-pro-preview",
     "standard-fallback": "openai/gpt-5-mini"
@@ -176558,6 +176599,7 @@ var init_model_config = __esm({
       "anthropic/claude-sonnet-4.6": { vision: true, reasoning: true },
       "anthropic/claude-sonnet-5": { vision: true, reasoning: true },
       "anthropic/claude-sonnet-5.5": { vision: true, reasoning: true },
+      "anthropic/claude-haiku-5.5": { 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 },
@@ -216468,6 +216510,12 @@ function getReasoningConfig(modelId, hasReasoningCapability = false, forceEnable
   const modelName = modelId.split("/")[1];
   switch (provider) {
     case "anthropic":
+      if (modelName.includes("claude-haiku-5")) {
+        return {
+          enabled: true,
+          providerOptions: { reasoning: { enabled: true, effort: reasoningEffort } }
+        };
+      }
       if (modelName.includes("claude-sonnet-4") || modelName.includes("claude-haiku-4") || modelName.includes("claude-4.5-sonnet") || modelName.includes("claude-4.5-haiku") || modelName.includes("claude-4") || modelName.includes("claude-3.7-sonnet") || modelName.includes("claude-3.5-sonnet-20241022")) {
         const effort = reasoningEffort || "high";
         const effortRatio = effort === "high" ? 0.8 : effort === "medium" ? 0.5 : 0.2;
@@ -220691,8 +220739,8 @@ HOW TO USE THIS:
 **Parameters**
 - \`description\`: one-line label that shows in the user's dock card (\u2264120 chars).
 - \`prompt\`: the actual work task. The subagent receives this as its user message.
-- \`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**.
+- \`model\`: **default to \`quick-thinking\` (alias for \`anthropic/claude-haiku-5.5\`) 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-5.5\`** (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-3.8-flash\`** (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-5.5\`** (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.
@@ -366428,14 +366476,19 @@ async function callLLM({
 
 ${systemMessage}` : systemMessage : claudeMessages.system;
     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 thinkingConfig = getClaudeThinkingConfig(
+      claudeModel,
+      reasoningConfig.enabled,
+      reasoningEffort,
+      reasoningOptions?.max_tokens
+    );
+    const legacyThinkingEnabled = thinkingConfig.thinking?.type === "enabled";
     const claudeRequest = {
       model: claudeModel,
       messages: claudeMessages.messages,
       system: mergedSystem,
-      max_tokens: maxTokens
+      max_tokens: maxTokens,
+      ...thinkingConfig
     };
     if (supportsSamplingParameters(claudeModel)) {
       claudeRequest.temperature = legacyThinkingEnabled ? 1 : temperature;
@@ -366443,19 +366496,6 @@ ${systemMessage}` : systemMessage : claudeMessages.system;
     if (shouldEnableTools && tools && tools.length > 0) {
       claudeRequest.tools = tools;
     }
-    if (thinkingEnabled) {
-      if (adaptiveThinkingEnabled) {
-        claudeRequest.thinking = { type: "adaptive" };
-        claudeRequest.output_config = {
-          effort: reasoningOptions?.effort ?? reasoningEffort ?? "high"
-        };
-      } else if (legacyThinkingEnabled && reasoningOptions?.max_tokens) {
-        claudeRequest.thinking = {
-          type: "enabled",
-          budget_tokens: reasoningOptions.max_tokens
-        };
-      }
-    }
     const leasedAttempt = beginLeasedProviderAttempt(
       claudeRoute,
       config2.bedrock.enabled ? configuredBedrockCircuitCredentialIdentity() : claudeApiKey,
@@ -393199,7 +393239,7 @@ function loadCliConfig(env2 = process.env) {
       // Haiku is a good CLI default: fast, cheap, large context. Any model works
       // now that the output-budget math accounts for the system prompt + tools
       // (see calculateMaxOutputTokens). Override with DEFAULT_MODEL or `-m`.
-      defaultModel: pick(env2.DEFAULT_MODEL, file2.defaultModel) || "anthropic/claude-haiku-4.5",
+      defaultModel: pick(env2.DEFAULT_MODEL, file2.defaultModel) || "anthropic/claude-haiku-5.5",
       maxSteps: 200,
       provider: pick(env2.USABLE_PROVIDER, file2.provider) ?? void 0,
       usableApiHost: pick(env2.USABLE_API_HOST, file2.usableApiHost),
@@ -393823,7 +393863,7 @@ init_tui_select();
 init_model_registry();
 
 // package.json
-var version2 = "1.215.12";
+var version2 = "1.216.0";
 
 // src/adapters/cli/model-catalog.ts
 init_codex_auth();
@@ -393901,7 +393941,7 @@ var PROVIDERS = [
     label: "Usable subscription \u2014 uses your login, no API key (recommended)",
     kind: "usable",
     provider: "usable",
-    defaultModel: "anthropic/claude-haiku-4.5"
+    defaultModel: "anthropic/claude-haiku-5.5"
   },
   {
     label: "OpenRouter \u2014 your API key, many models",
@@ -393918,7 +393958,7 @@ var PROVIDERS = [
     provider: "anthropic",
     keyField: "anthropic-api-key",
     keyHint: "sk-ant-\u2026",
-    defaultModel: "anthropic/claude-haiku-4.5"
+    defaultModel: "anthropic/claude-haiku-5.5"
   },
   {
     label: "OpenAI Codex \u2014 your ChatGPT subscription (sign in with ChatGPT)",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@usabledev/usable-chat",
-  "version": "1.215.12",
+  "version": "1.216.0",
   "description": "usable-chat — terminal harness for usable-chat (headless + TUI)",
   "type": "module",
   "bin": {