@ekkos/cli 1.5.0 → 1.5.2

package/dist/commands/dashboard.js

@@ -73,13 +73,21 @@ const MAX_OUTPUT_200K_OPUS_SONNET = 16384;
 // ── Pricing ──
 // Pricing per MTok from https://docs.anthropic.com/en/docs/about-claude/pricing
 // Includes prompt caching rates (5m cache write=1.25x input, cache read=0.1x input).
+// Pricing per MTok — source: https://platform.claude.com/docs/en/about-claude/pricing
+// Cache write = 5-minute TTL (1.25x input). 1-hour TTL (2x input) not tracked separately.
+// Fast mode (Opus 4.6 only, 6x standard) detected via x-ekkos-fast-mode header.
+// Data residency (US-only, 1.1x all categories) detected via inference_geo header.
 const MODEL_PRICING = {
+    // ── Opus family ──
     'claude-opus-4-6-20260514': { input: 5, output: 25, cacheWrite: 6.25, cacheRead: 0.50 },
     'claude-opus-4-6': { input: 5, output: 25, cacheWrite: 6.25, cacheRead: 0.50 },
+    'claude-opus-4-6-fast': { input: 30, output: 150, cacheWrite: 37.50, cacheRead: 3.00 }, // Fast mode (6x)
     'claude-opus-4-5-20251101': { input: 5, output: 25, cacheWrite: 6.25, cacheRead: 0.50 },
     'claude-opus-4-5': { input: 5, output: 25, cacheWrite: 6.25, cacheRead: 0.50 },
-    'claude-opus-4-20250514': { input: 5, output: 25, cacheWrite: 6.25, cacheRead: 0.50 },
-    'claude-opus-4': { input: 5, output: 25, cacheWrite: 6.25, cacheRead: 0.50 },
+    'claude-opus-4-1': { input: 15, output: 75, cacheWrite: 18.75, cacheRead: 1.50 },
+    'claude-opus-4-20250514': { input: 15, output: 75, cacheWrite: 18.75, cacheRead: 1.50 },
+    'claude-opus-4': { input: 15, output: 75, cacheWrite: 18.75, cacheRead: 1.50 },
+    // ── Sonnet family ──
     'claude-sonnet-4-6-20260514': { input: 3, output: 15, cacheWrite: 3.75, cacheRead: 0.30 },
     'claude-sonnet-4-6': { input: 3, output: 15, cacheWrite: 3.75, cacheRead: 0.30 },
     'claude-sonnet-4-5-20250929': { input: 3, output: 15, cacheWrite: 3.75, cacheRead: 0.30 },
@@ -87,8 +95,12 @@ const MODEL_PRICING = {
     'claude-sonnet-4-5-20250514': { input: 3, output: 15, cacheWrite: 3.75, cacheRead: 0.30 },
     'claude-sonnet-4-20250514': { input: 3, output: 15, cacheWrite: 3.75, cacheRead: 0.30 },
     'claude-sonnet-4': { input: 3, output: 15, cacheWrite: 3.75, cacheRead: 0.30 },
+    // ── Haiku family ──
     'claude-haiku-4-5-20251001': { input: 1, output: 5, cacheWrite: 1.25, cacheRead: 0.10 },
     'claude-haiku-4-5': { input: 1, output: 5, cacheWrite: 1.25, cacheRead: 0.10 },
+    'claude-haiku-3-5': { input: 0.80, output: 4, cacheWrite: 1.00, cacheRead: 0.08 },
+    'claude-haiku-3': { input: 0.25, output: 1.25, cacheWrite: 0.30, cacheRead: 0.03 },
+    // ── Gemini family ──
     'gemini-3.1-flash-lite-preview': { input: 0.05, output: 0.20, cacheWrite: 0, cacheRead: 0.05 },
     'gemini-3.1-pro-preview': { input: 1.25, output: 5.00, cacheWrite: 0, cacheRead: 1.25 },
     'gemini-3-pro-preview': { input: 1.25, output: 5.00, cacheWrite: 0, cacheRead: 1.25 },
@@ -100,25 +112,42 @@ const MODEL_PRICING = {
 function getModelPricing(modelId) {
     if (MODEL_PRICING[modelId])
         return MODEL_PRICING[modelId];
-    if (modelId.includes('gemini-3.1-pro'))
+    const lower = (modelId || '').toLowerCase();
+    // Gemini variants
+    if (lower.includes('gemini-3.1-pro'))
         return MODEL_PRICING['gemini-3.1-pro-preview'];
-    if (modelId.includes('gemini-3-pro'))
+    if (lower.includes('gemini-3-pro'))
         return MODEL_PRICING['gemini-3-pro-preview'];
-    if (modelId.includes('gemini-3.1-flash-lite'))
+    if (lower.includes('gemini-3.1-flash-lite'))
         return MODEL_PRICING['gemini-3.1-flash-lite-preview'];
-    if (modelId.includes('gemini-3-flash'))
+    if (lower.includes('gemini-3-flash'))
         return MODEL_PRICING['gemini-3-flash-preview'];
-    if (modelId.includes('gemini-2.5-pro'))
+    if (lower.includes('gemini-2.5-pro'))
         return MODEL_PRICING['gemini-2.5-pro'];
-    if (modelId.includes('gemini-2.5-flash-lite'))
+    if (lower.includes('gemini-2.5-flash-lite'))
         return MODEL_PRICING['gemini-2.5-flash-lite'];
-    if (modelId.includes('gemini-2.5-flash'))
+    if (lower.includes('gemini-2.5-flash'))
         return MODEL_PRICING['gemini-2.5-flash'];
-    if (modelId.includes('opus'))
+    // Claude Opus — version-aware (4.6/4.5 = $5, 4.1/4 = $15)
+    if (lower.includes('opus-4-6') || lower.includes('opus-4-5'))
         return MODEL_PRICING['claude-opus-4-6'];
-    if (modelId.includes('sonnet'))
+    if (lower.includes('opus-4-1') || lower.includes('opus-4-2'))
+        return MODEL_PRICING['claude-opus-4-1'];
+    if (lower.includes('opus-4'))
+        return MODEL_PRICING['claude-opus-4'];
+    if (lower.includes('opus'))
+        return MODEL_PRICING['claude-opus-4-6']; // Default to latest
+    // Claude Sonnet
+    if (lower.includes('sonnet'))
         return MODEL_PRICING['claude-sonnet-4-6'];
-    if (modelId.includes('haiku'))
+    // Claude Haiku — version-aware
+    if (lower.includes('haiku-4-5') || lower.includes('haiku-4'))
+        return MODEL_PRICING['claude-haiku-4-5'];
+    if (lower.includes('haiku-3-5'))
+        return MODEL_PRICING['claude-haiku-3-5'];
+    if (lower.includes('haiku-3'))
+        return MODEL_PRICING['claude-haiku-3'];
+    if (lower.includes('haiku'))
         return MODEL_PRICING['claude-haiku-4-5-20251001'];
     return MODEL_PRICING['claude-sonnet-4-6'];
 }
@@ -599,7 +628,8 @@ function buildDashboardData(params) {
         model: params.model,
         turnCount: turns.length,
         totalCost,
-        totalTokens: totalInput + totalCacheRead + totalCacheCreate + totalOutput,
+        // Unique content tokens: excludes cache_read (same prefix re-billed per turn)
+        totalTokens: totalInput + totalCacheCreate + totalOutput,
         totalInput,
         totalCacheRead,
         totalCacheCreate,
@@ -1915,7 +1945,12 @@ async function launchDashboard(initialSessionName, initialJsonlPath, refreshMs,
         if (!d)
             return;
         try {
-            const totalTokensM = ((d.totalInput + d.totalCacheRead + d.totalCacheCreate + d.totalOutput) / 1000000).toFixed(2);
+            // Show unique content tokens: output (model-generated) + cache_create (new content cached)
+            // + uncached input. Excludes cache_read (same prefix re-billed each turn — NOT unique content).
+            // Previously showed cumulative API total (input+read+create+output) which inflated by ~10-20x
+            // because the same prefix was counted on every tool round-trip.
+            const uniqueTokens = d.totalOutput + d.totalCacheCreate + d.totalInput;
+            const totalTokensM = (uniqueTokens / 1000000).toFixed(2);
             const totalSavings = d.turns.reduce((s, t) => s + t.savings, 0);
             // Model routing breakdown (uses routedModel for actual model counts)
             const opusCount = d.turns.filter(t => t.routedModel.includes('opus')).length;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ekkos/cli",
-  "version": "1.5.0",
+  "version": "1.5.2",
   "description": "ekkOS memory CLI — persistent memory for AI coding assistants (Claude Code, Gemini, Cursor, Windsurf)",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",