tokentracker-cli 0.91.0 → 0.91.3

package/src/lib/pricing/curated-overrides.json

@@ -2,7 +2,7 @@
   "_meta": {
     "note": "Curated price overrides. Always wins over LiteLLM. Two reasons to live here: (1) self-defined alias names that LiteLLM will never carry (kiro-*, hy3-*, composer-*, kimi-for-coding, free-tier OpenRouter routes); (2) prices we want to pin even if LiteLLM has the model (e.g. cache_write fields LiteLLM often omits). Units: USD per million tokens. Edit this file to override pricing without redeploying.",
     "units": "usd_per_million_tokens",
-    "deepseek_v4_pro_pricing_note": "The 2026-05-31 promotional-expiry warning is obsolete: api-docs.deepseek.com/quick_start/pricing verified on 2026-06-10 still lists input $0.435/M, output $0.87/M, cache_read $0.003625/M as the official v4-pro rates (the discount became the list price). Do NOT 'restore' the old 4x rates; re-verify against the official page before any change."
+    "deepseek_v4_pricing_note": "Official DeepSeek V4 pricing verified on 2026-08-21. Peak hours are 01:00-04:00 and 06:00-10:00 UTC; all other hours are off-peak at half price. Exact entries store peak rates. computeRowCost applies the time-of-use discount from each row's UTC hour_start."
   },
   "exact": {
     "claude-fable-5": {
@@ -133,16 +133,22 @@
       "cache_write": 0
     },
     "deepseek-v4-flash": {
-      "input": 0.14,
-      "output": 0.28,
-      "cache_read": 0.0028,
-      "cache_write": 0.14
+      "input": 0.44,
+      "output": 1.32,
+      "cache_read": 0.014,
+      "cache_write": 0.44
     },
     "deepseek-v4-pro": {
-      "input": 0.435,
-      "output": 0.87,
-      "cache_read": 0.003625,
-      "cache_write": 0.435
+      "input": 1.32,
+      "output": 3.96,
+      "cache_read": 0.044,
+      "cache_write": 1.32
+    },
+    "deepseek-v4-flash-vision-exp": {
+      "input": 0.44,
+      "output": 1.32,
+      "cache_read": 0.014,
+      "cache_write": 0.44
     },
     "deepseek-chat": {
       "input": 0.14,

package/src/lib/pricing/index.js

@@ -22,6 +22,10 @@ const PI_SUBSCRIPTION_SOURCES = new Set([
   "prime-agent-copilot",
 ]);
 const SEED_SNAPSHOT_PATH = path.resolve(__dirname, "seed-snapshot.json");
+const DEEPSEEK_TIME_PRICED_MODELS = [
+  "deepseek-v4-flash",
+  "deepseek-v4-pro",
+];
 
 // Sync seed load. Done at require-time so callers that haven't awaited
 // ensurePricingLoaded() (e.g. tests, vite mock startup, edge functions) still
@@ -117,6 +121,32 @@ function getModelPricing(model, opts = {}) {
   return ZERO_PRICING;
 }
 
+function isDeepSeekTimePricedModel(model) {
+  const lower = String(model || "").toLowerCase();
+  return DEEPSEEK_TIME_PRICED_MODELS.some((name) => lower.includes(name));
+}
+
+function isDeepSeekOffPeak(row) {
+  if (row?.pricing_tier === "off_peak") return true;
+  if (row?.pricing_tier === "peak") return false;
+  const timestamp = Date.parse(String(row?.hour_start || ""));
+  if (!Number.isFinite(timestamp)) return false;
+  const hour = new Date(timestamp).getUTCHours();
+  return !((hour >= 1 && hour < 4) || (hour >= 6 && hour < 10));
+}
+
+function getRowPricing(row) {
+  const pricing = getModelPricing(row?.model, { source: row?.source });
+  if (!isDeepSeekTimePricedModel(row?.model) || !isDeepSeekOffPeak(row)) return pricing;
+  return {
+    ...pricing,
+    input: (pricing.input || 0) * 0.5,
+    output: (pricing.output || 0) * 0.5,
+    cache_read: (pricing.cache_read || 0) * 0.5,
+    cache_write: (pricing.cache_write || 0) * 0.5,
+  };
+}
+
 // Same formula and Codex/every-code reasoning-folding rule as the previous
 // computeRowCost in src/lib/local-api.js. Moved here so vite mock + local
 // server share one source of truth.
@@ -125,7 +155,7 @@ function computeRowCost(row) {
   // usage record reports a zero marginal cost for those turns; do not
   // reinterpret the Claude model name as an Anthropic API bill.
   if (PI_SUBSCRIPTION_SOURCES.has(String(row?.source || "").toLowerCase())) return 0;
-  const pricing = getModelPricing(row.model, { source: row.source });
+  const pricing = getRowPricing(row);
   const reasoningIncludedInOutput = row.source === "codex" || row.source === "every-code";
   const reasoningCost = reasoningIncludedInOutput
     ? 0
@@ -152,6 +182,7 @@ module.exports = {
   ensurePricingLoaded,
   getPricingRevision,
   getModelPricing,
+  getRowPricing,
   computeRowCost,
   resetPricingForTests,
   MODEL_PRICING,