fix(stats): correct totalTokens calculation, provider pricing, and dashboard improvements

Author: k1yt

packages/types/src/providers/qwen-code.ts

@@ -10,8 +10,8 @@ export const qwenCodeModels = {
 		contextWindow: 1_000_000,
 		supportsImages: false,
 		supportsPromptCache: false,
-		inputPrice: 0,
-		outputPrice: 0,
+		inputPrice: 1.0,
+		outputPrice: 5.0,
 		cacheWritesPrice: 0,
 		cacheReadsPrice: 0,
 		description: "Qwen3 Coder Plus - High-performance coding model with 1M context window for large codebases",
@@ -21,8 +21,8 @@ export const qwenCodeModels = {
 		contextWindow: 1_000_000,
 		supportsImages: false,
 		supportsPromptCache: false,
-		inputPrice: 0,
-		outputPrice: 0,
+		inputPrice: 0.3,
+		outputPrice: 1.5,
 		cacheWritesPrice: 0,
 		cacheReadsPrice: 0,
 		description: "Qwen3 Coder Flash - Fast coding model with 1M context window optimized for speed",

src/api/providers/openai-codex.ts

@@ -8,6 +8,7 @@ import {
 	openAiCodexDefaultModelId,
 	OpenAiCodexModelId,
 	openAiCodexModels,
+	openAiNativeModels,
 	type ReasoningEffort,
 	type ReasoningEffortExtended,
 	ApiProviderError,
@@ -16,6 +17,7 @@ import { TelemetryService } from "@roo-code/telemetry"
 
 import { Package } from "../../shared/package"
 import type { ApiHandlerOptions } from "../../shared/api"
+import { calculateApiCostOpenAI } from "../../shared/cost"
 
 import { ApiStream, ApiStreamUsageChunk } from "../transform/stream"
 import { getModelParams } from "../transform/model-params"
@@ -126,15 +128,28 @@ export class OpenAiCodexHandler extends BaseProvider implements SingleCompletion
 				? usage.output_tokens_details.reasoning_tokens
 				: undefined
 
-		// Subscription-based: no per-token costs
+		// Compute equivalent API cost using openAiNativeModels pricing.
+		// The actual charge is covered by the ChatGPT Plus/Pro subscription,
+		// but showing the equivalent API cost lets users compare usage value.
+		const nativeModelInfo = openAiNativeModels[model.id as keyof typeof openAiNativeModels]
+		const { totalCost } = nativeModelInfo
+			? calculateApiCostOpenAI(
+					nativeModelInfo,
+					totalInputTokens,
+					totalOutputTokens,
+					cacheWriteTokens,
+					cacheReadTokens,
+				)
+			: { totalCost: 0 }
+
 		const out: ApiStreamUsageChunk = {
 			type: "usage",
 			inputTokens: totalInputTokens,
 			outputTokens: totalOutputTokens,
 			cacheWriteTokens,
 			cacheReadTokens,
 			...(typeof reasoningTokens === "number" ? { reasoningTokens } : {}),
-			totalCost: 0, // Subscription-based pricing
+			totalCost,
 		}
 		return out
 	}

src/services/stats/UsageAggregator.ts

@@ -70,11 +70,7 @@ export class UsageAggregator {
 	 * @param query Statistics query
 	 * @param options Additional options (e.g. recordingPaused)
 	 */
-	query(
-		events: UsageEventV1[],
-		query: StatsQuery,
-		options: { recordingPaused?: boolean } = {},
-	): StatsSnapshot {
+	query(events: UsageEventV1[], query: StatsQuery, options: { recordingPaused?: boolean } = {}): StatsSnapshot {
 		// 1. Time range filtering
 		const { from, to } = this.resolveTimeRange(query)
 		const filtered = events.filter((event) => {
@@ -86,9 +82,7 @@ export class UsageAggregator {
 
 		// 2. Cancelled event filtering
 		const includeCancelled = query.includeCancelled ?? false
-		const visibleEvents = includeCancelled
-			? filtered
-			: filtered.filter((e) => e.status !== "cancelled")
+		const visibleEvents = includeCancelled ? filtered : filtered.filter((e) => e.status !== "cancelled")
 
 		// 3. Compute bucket keys based on timezone
 		const aggregatable: AggregatableEvent[] = visibleEvents.map((event) => {
@@ -341,10 +335,7 @@ export class UsageAggregator {
 	 * Returns the bucket key combinations for the groupBy axes from the event.
 	 * Up to 3 axes can be combined.
 	 */
-	private getGroupKeys(
-		item: AggregatableEvent,
-		groupBy: StatsQuery["groupBy"],
-	): Record<string, string>[] {
+	private getGroupKeys(item: AggregatableEvent, groupBy: StatsQuery["groupBy"]): Record<string, string>[] {
 		if (groupBy.length === 0) {
 			return [{}]
 		}
@@ -397,33 +388,33 @@ export class UsageAggregator {
 			case "status":
 				return [event.status]
 			case "source": {
-					// Separate by the source of costUsd.
-					// Feature 1: If the event has no costUsd but the cost can be
-					// computed on-the-fly from model pricing, treat the source as
-					// "estimated" (since it is derived, not provider-reported).
-					const sources = new Set<string>()
-					if (event.usage.costUsd) {
-						sources.add(event.usage.costUsd.source)
-					} else {
-						// Check if cost can be computed; if so, mark as "estimated".
-						// Otherwise the source remains "unknown".
-						const computedCost = computeEventCost(event)
-						if (computedCost > 0) {
-							sources.add("estimated")
-						}
+				// Separate by the source of costUsd.
+				// Feature 1: If the event has no costUsd but the cost can be
+				// computed on-the-fly from model pricing, treat the source as
+				// "estimated" (since it is derived, not provider-reported).
+				const sources = new Set<string>()
+				if (event.usage.costUsd) {
+					sources.add(event.usage.costUsd.source)
+				} else {
+					// Check if cost can be computed; if so, mark as "estimated".
+					// Otherwise the source remains "unknown".
+					const computedCost = computeEventCost(event)
+					if (computedCost > 0) {
+						sources.add("estimated")
 					}
-					// Also consider the source of input/output tokens
-					if (event.usage.inputTokens) {
-						sources.add(event.usage.inputTokens.source)
-					}
-					if (event.usage.outputTokens) {
-						sources.add(event.usage.outputTokens.source)
-					}
-					if (sources.size === 0) {
-						sources.add("unknown")
-					}
-					return Array.from(sources)
 				}
+				// Also consider the source of input/output tokens
+				if (event.usage.inputTokens) {
+					sources.add(event.usage.inputTokens.source)
+				}
+				if (event.usage.outputTokens) {
+					sources.add(event.usage.outputTokens.source)
+				}
+				if (sources.size === 0) {
+					sources.add("unknown")
+				}
+				return Array.from(sources)
+			}
 			default:
 				return []
 		}
@@ -510,7 +501,9 @@ export class UsageAggregator {
 			bucket.reasoningTokens += reasoningTokens
 		}
 
-		bucket.totalTokens += totalTokens
+		// Recompute from input + output (provider-neutral) to repair historical events
+		// that may have been persisted with the old double-counted sum.
+		bucket.totalTokens += inputTokens + outputTokens
 		bucket.costUsd += costUsd
 	}
 
@@ -566,9 +559,7 @@ export class UsageAggregator {
 	): StatsSnapshot["coverage"] {
 		const times = visibleEvents.map((e) => new Date(e.event.occurredAt).getTime()).sort((a, b) => a - b)
 
-		const backfilledEventCount = visibleEvents.filter(
-			(e) => e.event.provenance === "history-backfill",
-		).length
+		const backfilledEventCount = visibleEvents.filter((e) => e.event.provenance === "history-backfill").length
 
 		return {
 			firstEventAt: times.length > 0 ? new Date(times[0]).toISOString() : undefined,

src/services/stats/UsageRecorder.ts

@@ -97,10 +97,8 @@ export class UsageRecorder {
 			model: ctx.model,
 			mode: ctx.mode,
 			usage: {
-				inputTokens:
-					ctx.inputTokens > 0 ? { value: ctx.inputTokens, source: ctx.tokenSource } : undefined,
-				outputTokens:
-					ctx.outputTokens > 0 ? { value: ctx.outputTokens, source: ctx.tokenSource } : undefined,
+				inputTokens: ctx.inputTokens > 0 ? { value: ctx.inputTokens, source: ctx.tokenSource } : undefined,
+				outputTokens: ctx.outputTokens > 0 ? { value: ctx.outputTokens, source: ctx.tokenSource } : undefined,
 				cacheWriteTokens: ctx.cacheWriteTokens
 					? { value: ctx.cacheWriteTokens, source: ctx.tokenSource }
 					: undefined,
@@ -110,19 +108,11 @@ export class UsageRecorder {
 				reasoningTokens: ctx.reasoningTokens
 					? { value: ctx.reasoningTokens, source: ctx.tokenSource }
 					: undefined,
-				// H3 fix: compute totalTokens at record time so aggregators/UI can rely on it.
-				// Sum all token buckets. Inclusion semantics (whether cache/reasoning are already
-				// counted inside input/output) are recorded in `semantics` below; the aggregator
-				// is responsible for adjusting double-counting when semantics != "unknown".
-				// Until provider-specific semantics are determined, we record the raw sum so the
-				// total is never 0 (which previously broke heatmap/sort).
+				// totalTokens = inputTokens + outputTokens (provider-neutral definition).
+				// Cache tokens are a subset/breakdown of input; reasoning tokens are a subset of output.
+				// Adding them separately would double-count. See docs/260720_22_gitignore-heatmap-fix/213200_debug-report.md
 				totalTokens: {
-					value:
-						ctx.inputTokens +
-						ctx.outputTokens +
-						(ctx.cacheReadTokens ?? 0) +
-						(ctx.cacheWriteTokens ?? 0) +
-						(ctx.reasoningTokens ?? 0),
+					value: ctx.inputTokens + ctx.outputTokens,
 					source: ctx.tokenSource,
 				},
 				costUsd: ctx.totalCost ? { value: ctx.totalCost, source: ctx.costSource } : undefined,