codeep 3.3.1 → 3.3.2

package/dist/renderer/commands/helpers.d.ts

@@ -85,6 +85,10 @@ export interface StatsCache {
     cacheReadTokens: number;
     cacheCreationTokens: number;
     estimatedSavingsUsd: number;
+    /** Rates that applied, from getCacheStats. Absent → no rate is quoted rather
+     *  than assuming Anthropic's 0.1×, which was wrong for DeepSeek, Kimi, Qwen
+     *  and Fable 5.1. */
+    cacheReadRates?: number[];
 }
 export interface PricingRow {
     model: string;

package/dist/renderer/commands/helpers.js

@@ -8,6 +8,7 @@
  * coverage.
  */
 import { isFlatFeeProvider } from '../../config/providers.js';
+import { formatCacheReadRates } from '../../utils/tokenTracker.js';
 /** Snippet window: chars of context before / after the match. */
 export const SEARCH_SNIPPET_BEFORE = 30;
 export const SEARCH_SNIPPET_AFTER = 50;
@@ -166,7 +167,7 @@ export function formatStatsReport(args) {
         }
         if (cache.cacheReadTokens > 0 || cache.cacheCreationTokens > 0) {
             lines.push('', '### Prompt caching');
-            lines.push(`Cache reads: ${fmt(cache.cacheReadTokens)} tokens (billed at 0.1× input rate)`);
+            lines.push(`Cache reads: ${fmt(cache.cacheReadTokens)} tokens${formatCacheReadRates(cache.cacheReadRates ?? [])}`);
             if (cache.cacheCreationTokens > 0) {
                 lines.push(`Cache writes: ${fmt(cache.cacheCreationTokens)} tokens (billed at 1.25× input rate)`);
             }

package/dist/utils/tokenTracker.d.ts

@@ -93,6 +93,21 @@ export interface ProviderCostBreakdown {
     cacheReadTokens: number;
     estimatedCost: number;
 }
+/**
+ * What a cached prompt token costs, as a fraction of the model's input rate.
+ * Model first (a property of the model), then provider, then the default.
+ *
+ * One lookup for everything that needs it. Cost used this chain while savings
+ * hardcoded 0.1, so every provider priced differently from Anthropic had a
+ * cost and a "saved" figure that disagreed with each other.
+ */
+export declare function cacheReadRateFor(model: string, provider: string | undefined): number;
+/**
+ * The rate note for a report. One rate reads as "0.02×"; a session mixing
+ * providers reads as a range, because any single number there would be wrong
+ * for part of it.
+ */
+export declare function formatCacheReadRates(rates: readonly number[]): string;
 /**
  * Get cost breakdown grouped by provider/model.
  *
@@ -120,6 +135,9 @@ export interface CacheStats {
     /** True when EVERY cached token came from a flat-fee plan — there is no
      *  metered spend to have saved against. */
     isEntirelyFlatFeeCache: boolean;
+    /** The read rate of each metered record that read from cache, so a report
+     *  can state the rate that actually applied instead of assuming 0.1×. */
+    cacheReadRates: number[];
 }
 export declare function getCacheStats(): CacheStats;
 /**

package/dist/utils/tokenTracker.js

@@ -127,9 +127,10 @@ const MODEL_PRICING = {
     // rate, so it carries PEAK — an over-estimate by design. The previous rows
     // (0.435/0.87 and 0.14/0.28) were two to four and a half times below today's
     // peak and so under-reported, which is the one direction this table must not
-    // err in. Cache hits cost 2% of a miss, but DeepSeek reports them as
-    // `prompt_cache_hit_tokens`, which this tracker does not read — so every input
-    // token bills at the miss rate: an over-estimate again, not a gap.
+    // err in. Cache hits are read (DeepSeek reports them both nested as
+    // `prompt_tokens_details.cached_tokens` and top-level as
+    // `prompt_cache_hit_tokens`) and billed at DeepSeek's own hit rate — see
+    // CACHE_READ_RATE below.
     'deepseek-flash': { inputPer1M: 0.30, outputPer1M: 1.20 },
     // Retired V4 Flash is served by V4.1 Flash and billed at its price.
     'deepseek-v4-flash': { inputPer1M: 0.30, outputPer1M: 1.20 },
@@ -250,8 +251,12 @@ export function extractOpenAIUsage(data) {
         // later. Reading only the nested form zeroed every Kimi cache hit, so the
         // cached portion of a run billed at the full cache-miss rate — five times
         // what it costs — with nothing anywhere to say so.
+        // DeepSeek sends the same number twice — nested `cached_tokens` and
+        // top-level `prompt_cache_hit_tokens` — so the nested read already covers
+        // it; the top-level field is a last resort, never added to the other.
         const nested = data.usage.prompt_tokens_details?.cached_tokens;
-        const cached = (typeof nested === 'number' ? nested : data.usage.cached_tokens) || 0;
+        const topLevel = data.usage.cached_tokens ?? data.usage.prompt_cache_hit_tokens;
+        const cached = (typeof nested === 'number' ? nested : topLevel) || 0;
         return {
             promptTokens: data.usage.prompt_tokens || 0,
             completionTokens: data.usage.completion_tokens || 0,
@@ -301,8 +306,16 @@ export function extractAnthropicUsage(data) {
  */
 const MODEL_CACHE_READ_RATE = {
     'claude-fable-5-1': 0.025,
+    // V4 Pro's own ratio: $0.044 hit against $1.32 miss (peak; off-peak halves
+    // both, so the ratio holds). Historical — V4 Pro routes to V4.1 Flash from
+    // 2026-09-14 and configs holding it are migrated.
+    'deepseek-v4-pro': 0.044 / 1.32,
 };
 const CACHE_READ_RATE = {
+    // V4.1 Flash: $0.006 hit against $0.30 miss at peak, $0.003 against $0.15
+    // off-peak — 0.02 either way. Without an entry DeepSeek fell to the 0.1
+    // default and every cached token billed at five times its price.
+    'deepseek': 0.02,
     'kimi': 0.2,
     'kimi-api': 0.2,
     'qwen': 0.2,
@@ -313,6 +326,33 @@ const CACHE_READ_RATE = {
 };
 /** Anthropic's ratio, and the safest guess for a provider we have not priced. */
 const DEFAULT_CACHE_READ_RATE = 0.1;
+/**
+ * What a cached prompt token costs, as a fraction of the model's input rate.
+ * Model first (a property of the model), then provider, then the default.
+ *
+ * One lookup for everything that needs it. Cost used this chain while savings
+ * hardcoded 0.1, so every provider priced differently from Anthropic had a
+ * cost and a "saved" figure that disagreed with each other.
+ */
+export function cacheReadRateFor(model, provider) {
+    return MODEL_CACHE_READ_RATE[model]
+        ?? CACHE_READ_RATE[provider?.trim().toLowerCase() ?? '']
+        ?? DEFAULT_CACHE_READ_RATE;
+}
+/**
+ * The rate note for a report. One rate reads as "0.02×"; a session mixing
+ * providers reads as a range, because any single number there would be wrong
+ * for part of it.
+ */
+export function formatCacheReadRates(rates) {
+    const fmt = (r) => `${Number(r.toFixed(3))}×`;
+    const distinct = [...new Set(rates.map(r => Number(r.toFixed(4))))].sort((a, b) => a - b);
+    if (distinct.length === 0)
+        return '';
+    if (distinct.length === 1)
+        return ` (billed at ${fmt(distinct[0])} input rate)`;
+    return ` (billed at ${fmt(distinct[0])}–${fmt(distinct[distinct.length - 1])} input rate, by model)`;
+}
 /**
  * Get cost breakdown grouped by provider/model.
  *
@@ -346,9 +386,7 @@ export function getCostBreakdown(startIndex = 0) {
                 // prompt tokens bill at the standard 1.0× rate.
                 const cacheCreate = record.cacheCreationTokens ?? 0;
                 const cacheRead = record.cacheReadTokens ?? 0;
-                const cacheReadRate = MODEL_CACHE_READ_RATE[record.model]
-                    ?? CACHE_READ_RATE[record.provider?.trim().toLowerCase()]
-                    ?? DEFAULT_CACHE_READ_RATE;
+                const cacheReadRate = cacheReadRateFor(record.model, record.provider);
                 const uncachedPrompt = Math.max(0, record.promptTokens - cacheCreate - cacheRead);
                 existing.estimatedCost +=
                     (uncachedPrompt / 1_000_000) * pricing.inputPer1M
@@ -367,6 +405,7 @@ export function getCacheStats() {
     let savings = 0;
     let flatFeeCached = 0;
     let meteredCached = 0;
+    const readRates = [];
     for (const record of currentRecords()) {
         const cached = (record.cacheCreationTokens ?? 0) + (record.cacheReadTokens ?? 0);
         cacheCreate += record.cacheCreationTokens ?? 0;
@@ -380,11 +419,16 @@ export function getCacheStats() {
         }
         meteredCached += cached;
         // Savings = what cache-read tokens would have cost at full input rate,
-        // minus what they actually cost at 0.1×. (Cache creation is a slight
-        // *penalty* of 0.25× — netted in for honest reporting.)
+        // minus what they cost at the model's own read rate. This hardcoded 0.9
+        // (a 0.1 read) for every provider, so Kimi and Qwen (0.2) over-reported
+        // savings while DeepSeek (0.02) and Fable 5.1 (0.025) under-reported them.
+        // (Cache creation is a slight *penalty* of 0.25× — netted in.)
         const pricing = MODEL_PRICING[record.model];
         if (pricing) {
-            const cReadSaved = ((record.cacheReadTokens ?? 0) / 1_000_000) * pricing.inputPer1M * 0.9;
+            const readRate = cacheReadRateFor(record.model, record.provider);
+            if ((record.cacheReadTokens ?? 0) > 0)
+                readRates.push(readRate);
+            const cReadSaved = ((record.cacheReadTokens ?? 0) / 1_000_000) * pricing.inputPer1M * (1 - readRate);
             const cCreateCost = ((record.cacheCreationTokens ?? 0) / 1_000_000) * pricing.inputPer1M * 0.25;
             savings += cReadSaved - cCreateCost;
         }
@@ -395,6 +439,7 @@ export function getCacheStats() {
         estimatedSavingsUsd: Math.max(0, savings),
         hasFlatFeeCacheUsage: flatFeeCached > 0,
         isEntirelyFlatFeeCache: flatFeeCached > 0 && meteredCached === 0,
+        cacheReadRates: readRates,
     };
 }
 /**
@@ -506,7 +551,7 @@ export function formatCostReport() {
         // The billing multipliers only describe a metered account. On a plan
         // nothing is billed per token, so quoting a rate there would be as invented
         // as the per-model prices this report already refuses to show.
-        const readNote = cache.isEntirelyFlatFeeCache ? '' : ' (billed at 0.1× input rate)';
+        const readNote = cache.isEntirelyFlatFeeCache ? '' : formatCacheReadRates(cache.cacheReadRates);
         const writeNote = cache.isEntirelyFlatFeeCache ? '' : ' (billed at 1.25× input rate)';
         lines.push(`**Cache reads:** ${formatTokenCount(cache.cacheReadTokens)} tokens${readNote}`);
         if (cache.cacheCreationTokens > 0) {

package/dist/version.d.ts

@@ -1 +1 @@
-export declare const VERSION = "3.3.1";
+export declare const VERSION = "3.3.2";

package/dist/version.js

@@ -1,4 +1,4 @@
 // AUTO-GENERATED by scripts/gen-version.js — do not edit by hand.
 // Baked from package.json at build time so the bun-compiled binary reports
 // the right version (it has no package.json on disk to read at runtime).
-export const VERSION = '3.3.1';
+export const VERSION = '3.3.2';

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codeep",
-  "version": "3.3.1",
+  "version": "3.3.2",
   "description": "AI-powered coding assistant built for the terminal. Multiple LLM providers, project-aware context, and a seamless development workflow.",
   "type": "module",
   "main": "dist/index.js",