Skip to content

The open alternative to Salesforce, designed for AI.

fix(ai): price multi-step turns step by step (#26882)

Etienne8c8f3e12ff3e
diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-agent-execution/services/agent-async-executor.service.ts b/packages/twenty-server/src/engine/metadata-modules/ai/ai-agent-execution/services/agent-async-executor.service.ts
index 0f24579585d..a8f719c470a 100644
--- a/packages/twenty-server/src/engine/metadata-modules/ai/ai-agent-execution/services/agent-async-executor.service.ts
+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-agent-execution/services/agent-async-executor.service.ts
@@ -566,9 +566,9 @@ export class AgentAsyncExecutorService {
result = structuredResult.output as object;
}
- const tokenCostInDollars = this.aiBillingService.calculateCost(
+ const tokenCostInDollars = this.aiBillingService.calculateStepsCost(
registeredModel.modelId,
- { usage: accumulatedUsage, cacheCreationTokens },
+ executionSteps,
);
const totalCostInDollars =
tokenCostInDollars +
@@ -602,10 +602,10 @@ export class AgentAsyncExecutorService {
// Nothing was generated when execution failed before a model resolved,
// and pricing an unresolved id would throw over the original error.
const costInDollars = isDefined(resolvedModelId)
- ? this.aiBillingService.calculateCost(resolvedModelId, {
- usage: accumulatedUsage,
- cacheCreationTokens,
- })
+ ? this.aiBillingService.calculateStepsCost(
+ resolvedModelId,
+ executionSteps,
+ )
: 0;
const creditsUsedMicro = convertDollarsToCreditsMicro(costInDollars);
const totalTokens =
diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/services/ai-billing.service.ts b/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/services/ai-billing.service.ts
index 5a101505d23..fcd798588d5 100644
--- a/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/services/ai-billing.service.ts
+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/services/ai-billing.service.ts
@@ -10,8 +10,9 @@ import { UsageRecorderService } from 'src/engine/core-modules/usage/services/usa
import { type UsageSpenders } from 'src/engine/core-modules/usage/types/usage-spenders.type';
import { NATIVE_WEB_SEARCH_COST_PER_CALL_DOLLARS } from 'src/engine/metadata-modules/ai/ai-billing/constants/native-web-search-cost-per-call-dollars';
import { type BillingTokenUsage } from 'src/engine/metadata-modules/ai/ai-billing/types/billing-token-usage.type';
-import { computeCostBreakdown } from 'src/engine/metadata-modules/ai/ai-billing/utils/compute-cost-breakdown.util';
+import { computeStepCostBreakdown } from 'src/engine/metadata-modules/ai/ai-billing/utils/compute-step-cost-breakdown.util';
import { convertDollarsToCreditsMicro } from 'src/engine/metadata-modules/ai/ai-billing/utils/convert-dollars-to-credits-micro.util';
+import { extractCacheCreationTokens } from 'src/engine/metadata-modules/ai/ai-billing/utils/extract-cache-creation-tokens.util';
import { AiModelRegistryService } from 'src/engine/metadata-modules/ai/ai-models/services/ai-model-registry.service';
import { type AiModelCostConfig } from 'src/engine/metadata-modules/ai/ai-models/types/ai-model-cost-config.type';
import { type ModelId } from 'src/engine/metadata-modules/ai/ai-models/types/model-id.type';
@@ -85,17 +86,31 @@ export class AiBillingService {
);
}
+ calculateStepsCost(
+ modelId: ModelId,
+ steps: {
+ usage: BillingTokenUsage;
+ providerMetadata?: Record<string, Record<string, unknown> | undefined>;
+ }[],
+ ): number {
+ const model = this.getCostConfig(modelId);
+
+ return steps.reduce(
+ (costInDollars, step) =>
+ costInDollars +
+ computeStepCostBreakdown(model, {
+ usage: step.usage,
+ cacheCreationTokens: extractCacheCreationTokens(
+ step.providerMetadata,
+ ),
+ }).totalCostInDollars,
+ 0,
+ );
+ }
+
calculateCost(modelId: ModelId, billingInput: BillingUsageInput): number {
const model = this.getCostConfig(modelId);
- const { usage, cacheCreationTokens = 0 } = billingInput;
-
- const breakdown = computeCostBreakdown(model, {
- inputTokens: usage.inputTokens,
- outputTokens: usage.outputTokens,
- reasoningTokens: usage.outputTokenDetails?.reasoningTokens,
- cachedInputTokens: usage.inputTokenDetails?.cacheReadTokens,
- cacheCreationTokens,
- });
+ const breakdown = computeStepCostBreakdown(model, billingInput);
this.logger.log(
`Cost for ${model.modelId}: $${breakdown.totalCostInDollars.toFixed(6)} ` +
diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/utils/__tests__/compute-step-cost-breakdown.util.spec.ts b/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/utils/__tests__/compute-step-cost-breakdown.util.spec.ts
new file mode 100644
index 00000000000..b88d513a271
--- /dev/null
+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/utils/__tests__/compute-step-cost-breakdown.util.spec.ts
@@ -0,0 +1,65 @@
+import { computeStepCostBreakdown } from 'src/engine/metadata-modules/ai/ai-billing/utils/compute-step-cost-breakdown.util';
+import { type AiModelCostConfig } from 'src/engine/metadata-modules/ai/ai-models/types/ai-model-cost-config.type';
+
+const MODEL_COST_CONFIG: AiModelCostConfig = {
+ modelId: 'test-model',
+ inputCostPerMillionTokens: 1,
+ outputCostPerMillionTokens: 2,
+ cachedInputCostPerMillionTokens: 0.5,
+ cacheCreationCostPerMillionTokens: 3,
+ longContextCost: {
+ inputCostPerMillionTokens: 10,
+ outputCostPerMillionTokens: 20,
+ cacheCreationCostPerMillionTokens: 30,
+ thresholdTokens: 200_000,
+ },
+};
+
+describe('computeStepCostBreakdown', () => {
+ it('applies the long-context rate only when the step itself crosses the threshold', () => {
+ const stepUnderThreshold = computeStepCostBreakdown(MODEL_COST_CONFIG, {
+ usage: { inputTokens: 150_000, outputTokens: 1_000 },
+ });
+ const stepOverThreshold = computeStepCostBreakdown(MODEL_COST_CONFIG, {
+ usage: { inputTokens: 300_000, outputTokens: 2_000 },
+ });
+
+ expect(2 * stepUnderThreshold.totalCostInDollars).toBeCloseTo(0.304);
+ expect(stepOverThreshold.totalCostInDollars).toBeCloseTo(3.04);
+ });
+
+ it('reads cache reads and reasoning tokens from the usage details', () => {
+ const breakdown = computeStepCostBreakdown(MODEL_COST_CONFIG, {
+ usage: {
+ inputTokens: 100_000,
+ outputTokens: 10_000,
+ inputTokenDetails: { cacheReadTokens: 40_000 },
+ outputTokenDetails: { reasoningTokens: 4_000 },
+ },
+ });
+
+ expect(breakdown.inputCostInDollars).toBeCloseTo(0.08);
+ expect(breakdown.outputCostInDollars).toBeCloseTo(0.02);
+ expect(breakdown.tokenCounts).toMatchObject({
+ adjustedInputTokens: 60_000,
+ cachedInputTokens: 40_000,
+ reasoningTokens: 4_000,
+ });
+ });
+
+ it('prices cache creation tokens at the cache creation rate', () => {
+ const breakdown = computeStepCostBreakdown(MODEL_COST_CONFIG, {
+ usage: { inputTokens: 100_000, outputTokens: 0 },
+ cacheCreationTokens: 100_000,
+ });
+
+ expect(breakdown.totalCostInDollars).toBeCloseTo(0.3);
+ expect(breakdown.tokenCounts.adjustedInputTokens).toBe(0);
+ });
+
+ it('costs nothing for a step without usage', () => {
+ expect(
+ computeStepCostBreakdown(MODEL_COST_CONFIG, {}).totalCostInDollars,
+ ).toBe(0);
+ });
+});
diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/utils/compute-step-cost-breakdown.util.ts b/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/utils/compute-step-cost-breakdown.util.ts
new file mode 100644
index 00000000000..4fea2f5c84c
--- /dev/null
+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-billing/utils/compute-step-cost-breakdown.util.ts
@@ -0,0 +1,21 @@
+import { type BillingTokenUsage } from 'src/engine/metadata-modules/ai/ai-billing/types/billing-token-usage.type';
+import {
+ computeCostBreakdown,
+ type CostBreakdown,
+} from 'src/engine/metadata-modules/ai/ai-billing/utils/compute-cost-breakdown.util';
+import { type AiModelCostConfig } from 'src/engine/metadata-modules/ai/ai-models/types/ai-model-cost-config.type';
+
+export const computeStepCostBreakdown = (
+ model: AiModelCostConfig,
+ {
+ usage,
+ cacheCreationTokens = 0,
+ }: { usage?: BillingTokenUsage; cacheCreationTokens?: number },
+): CostBreakdown =>
+ computeCostBreakdown(model, {
+ inputTokens: usage?.inputTokens,
+ outputTokens: usage?.outputTokens,
+ reasoningTokens: usage?.outputTokenDetails?.reasoningTokens,
+ cachedInputTokens: usage?.inputTokenDetails?.cacheReadTokens,
+ cacheCreationTokens,
+ });
diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/jobs/__tests__/stream-agent-chat.job.spec.ts b/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/jobs/__tests__/stream-agent-chat.job.spec.ts
index 2d3c5dbe619..3cf1cd8197a 100644
--- a/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/jobs/__tests__/stream-agent-chat.job.spec.ts
+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/jobs/__tests__/stream-agent-chat.job.spec.ts
@@ -5,6 +5,7 @@ import { isDefined } from 'twenty-shared/utils';
import { type WorkspaceEntity } from 'src/engine/core-modules/workspace/workspace.entity';
import { StreamAgentChatJob } from 'src/engine/metadata-modules/ai/ai-chat/jobs/stream-agent-chat.job';
import { type StreamAgentChatJobData } from 'src/engine/metadata-modules/ai/ai-chat/jobs/stream-agent-chat-job.types';
+import { type AiModelConfig } from 'src/engine/metadata-modules/ai/ai-models/types/ai-model-config.type';
import { AiExceptionCode } from 'src/engine/metadata-modules/ai/ai.exception';
type PublishedEvent = { type: string } & Record<string, unknown>;
@@ -28,35 +29,39 @@ const START_PARTS: ModelStreamPart[] = [
{ type: 'start-step', request: {}, warnings: [] },
];
-const FINISH_PARTS: ModelStreamPart[] = [
- {
- type: 'finish-step',
- response: {
- id: 'response-id',
- timestamp: new Date(0),
- modelId: 'gpt-5.6-luna',
- },
- usage: USAGE,
- performance: {
- effectiveOutputTokensPerSecond: 0,
- outputTokensPerSecond: undefined,
- inputTokensPerSecond: undefined,
- effectiveTotalTokensPerSecond: 0,
- stepTimeMs: 0,
- responseTimeMs: 0,
- timeToFirstOutputMs: 0,
- toolExecutionMs: {},
- },
- finishReason: 'stop',
- rawFinishReason: undefined,
- providerMetadata: undefined,
+const buildFinishStepPart = (usage: LanguageModelUsage): ModelStreamPart => ({
+ type: 'finish-step',
+ response: {
+ id: 'response-id',
+ timestamp: new Date(0),
+ modelId: 'gpt-5.6-luna',
},
- {
- type: 'finish',
- finishReason: 'stop',
- rawFinishReason: undefined,
- totalUsage: USAGE,
+ usage,
+ performance: {
+ effectiveOutputTokensPerSecond: 0,
+ outputTokensPerSecond: undefined,
+ inputTokensPerSecond: undefined,
+ effectiveTotalTokensPerSecond: 0,
+ stepTimeMs: 0,
+ responseTimeMs: 0,
+ timeToFirstOutputMs: 0,
+ toolExecutionMs: {},
},
+ finishReason: 'stop',
+ rawFinishReason: undefined,
+ providerMetadata: undefined,
+});
+
+const buildFinishPart = (totalUsage: LanguageModelUsage): ModelStreamPart => ({
+ type: 'finish',
+ finishReason: 'stop',
+ rawFinishReason: undefined,
+ totalUsage,
+});
+
+const FINISH_PARTS: ModelStreamPart[] = [
+ buildFinishStepPart(USAGE),
+ buildFinishPart(USAGE),
];
const TEXT_PARTS: ModelStreamPart[] = [
@@ -159,6 +164,11 @@ describe('StreamAgentChatJob', () => {
assistantPersistRejection,
totalsUpdateAffected = 1,
finalPublishRejection,
+ modelConfig = {
+ contextWindowTokens: 100000,
+ inputCostPerMillionTokens: 1,
+ outputCostPerMillionTokens: 2,
+ },
}: {
workspaceFound?: boolean;
chatStream?: ReturnType<typeof createFakeChatStream>;
@@ -166,9 +176,15 @@ describe('StreamAgentChatJob', () => {
assistantPersistRejection?: Error;
totalsUpdateAffected?: number;
finalPublishRejection?: Error;
+ modelConfig?: Partial<AiModelConfig>;
} = {}) => {
const publishedEvents: PublishedEvent[] = [];
+ const threadUsageQuery = jest.fn(async () =>
+ Array.from({ length: totalsUpdateAffected }, () => ({
+ id: 'thread-id',
+ })),
+ );
const threadRepository = {
findOne: jest.fn().mockResolvedValue({
id: 'thread-id',
@@ -180,12 +196,7 @@ describe('StreamAgentChatJob', () => {
work({
storage: 'core',
table: () => 'core."agentChatThread"',
- manager: {
- query: async () =>
- Array.from({ length: totalsUpdateAffected }, () => ({
- id: 'thread-id',
- })),
- },
+ manager: { query: threadUsageQuery },
}),
),
};
@@ -205,11 +216,7 @@ describe('StreamAgentChatJob', () => {
? jest.fn().mockRejectedValue(streamChatRejection)
: jest.fn().mockResolvedValue({
stream: chatStream,
- modelConfig: {
- contextWindowTokens: 100000,
- inputCostPerMillionTokens: 1,
- outputCostPerMillionTokens: 2,
- },
+ modelConfig,
hasNoMoreAvailableCredits: () => false,
}),
};
@@ -286,6 +293,7 @@ describe('StreamAgentChatJob', () => {
chatExecutionService,
publishedEvents,
threadRepository,
+ threadUsageQuery,
agentChatService,
eventPublisherService,
agentChatStreamingService,
@@ -402,6 +410,67 @@ describe('StreamAgentChatJob', () => {
expect(agentChatService.notifyThreadUsageUpdated).not.toHaveBeenCalled();
});
+ it('prices each step on its own, so steps under the long-context threshold never reach its rate together', async () => {
+ const stepUsage: LanguageModelUsage = {
+ inputTokens: 150_000,
+ inputTokenDetails: {
+ noCacheTokens: 150_000,
+ cacheReadTokens: 0,
+ cacheWriteTokens: 0,
+ },
+ outputTokens: 1_000,
+ outputTokenDetails: { textTokens: 1_000, reasoningTokens: 0 },
+ totalTokens: 151_000,
+ };
+ const turnUsage: LanguageModelUsage = {
+ inputTokens: 300_000,
+ inputTokenDetails: {
+ noCacheTokens: 300_000,
+ cacheReadTokens: 0,
+ cacheWriteTokens: 0,
+ },
+ outputTokens: 2_000,
+ outputTokenDetails: { textTokens: 2_000, reasoningTokens: 0 },
+ totalTokens: 302_000,
+ };
+ const { job, threadUsageQuery } = buildJob({
+ chatStream: createFakeChatStream({
+ parts: [
+ ...START_PARTS,
+ { type: 'text-start', id: 'text-1' },
+ { type: 'text-delta', id: 'text-1', text: 'Hello' },
+ { type: 'text-end', id: 'text-1' },
+ buildFinishStepPart(stepUsage),
+ { type: 'start-step', request: {}, warnings: [] },
+ buildFinishStepPart(stepUsage),
+ buildFinishPart(turnUsage),
+ ],
+ }),
+ modelConfig: {
+ contextWindowTokens: 1_000_000,
+ inputCostPerMillionTokens: 1,
+ outputCostPerMillionTokens: 2,
+ longContextCost: {
+ inputCostPerMillionTokens: 10,
+ outputCostPerMillionTokens: 20,
+ thresholdTokens: 200_000,
+ },
+ },
+ });
+
+ await job.handle(jobData);
+
+ const [, [, , inputTokens, outputTokens, inputCredits, outputCredits]] =
+ threadUsageQuery.mock.calls[0] as unknown as [string, number[]];
+
+ expect({ inputTokens, outputTokens, inputCredits, outputCredits }).toEqual({
+ inputTokens: 300_000,
+ outputTokens: 2_000,
+ inputCredits: 300_000,
+ outputCredits: 4_000,
+ });
+ });
+
it('applies thread totals when the claim is still held even if the message already exists from a checkpoint', async () => {
const { job, agentChatService } = buildJob({ totalsUpdateAffected: 1 });
@@ -766,6 +835,31 @@ describe('StreamAgentChatJob', () => {
expect(turnCounts('ai-chat/turn-completed')).toEqual([]);
});
+ it('adds the steps an aborted turn completed to the thread totals', async () => {
+ const { job, threadUsageQuery } = buildJob({
+ chatStream: createFakeChatStream({
+ parts: [
+ ...START_PARTS,
+ { type: 'text-start', id: 'text-1' },
+ { type: 'text-delta', id: 'text-1', text: 'Hello' },
+ { type: 'text-end', id: 'text-1' },
+ buildFinishStepPart(USAGE),
+ ],
+ isAborted: true,
+ }),
+ });
+
+ await job.handle(jobData);
+
+ const [, [, , inputTokens, outputTokens]] = threadUsageQuery.mock
+ .calls[0] as unknown as [string, number[]];
+
+ expect({ inputTokens, outputTokens }).toEqual({
+ inputTokens: 12,
+ outputTokens: 3,
+ });
+ });
+
it('counts a turn whose claim moved on as superseded', async () => {
const { job, turnCounts } = buildJob({ totalsUpdateAffected: 0 });
diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/jobs/stream-agent-chat.job.ts b/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/jobs/stream-agent-chat.job.ts
index 9aa6a7fb8e7..e593a58871d 100644
--- a/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/jobs/stream-agent-chat.job.ts
+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/jobs/stream-agent-chat.job.ts
@@ -28,7 +28,7 @@ import { MetricsService } from 'src/engine/core-modules/metrics/metrics.service'
import { MetricsKeys } from 'src/engine/core-modules/metrics/types/metrics-keys.type';
import { toDisplayCredits } from 'src/engine/core-modules/usage/utils/to-display-credits.util';
import { WorkspaceEntity } from 'src/engine/core-modules/workspace/workspace.entity';
-import { computeCostBreakdown } from 'src/engine/metadata-modules/ai/ai-billing/utils/compute-cost-breakdown.util';
+import { computeStepCostBreakdown } from 'src/engine/metadata-modules/ai/ai-billing/utils/compute-step-cost-breakdown.util';
import { convertDollarsToCreditsMicro } from 'src/engine/metadata-modules/ai/ai-billing/utils/convert-dollars-to-credits-micro.util';
import { extractCacheCreationTokens } from 'src/engine/metadata-modules/ai/ai-billing/utils/extract-cache-creation-tokens.util';
import {
@@ -505,6 +505,7 @@ export class StreamAgentChatJob {
return this.computeMessageMetadata({
part,
modelConfig,
+ streamUsage,
lastStepConversationSize,
totalCacheCreationTokens,
onUpdateUsage: (usage) => {
@@ -653,6 +654,7 @@ export class StreamAgentChatJob {
private computeMessageMetadata({
part,
modelConfig,
+ streamUsage,
lastStepConversationSize,
totalCacheCreationTokens,
onUpdateUsage,
@@ -663,9 +665,6 @@ export class StreamAgentChatJob {
type: string;
usage?: {
inputTokens?: number;
- };
- totalUsage?: {
- inputTokens?: number;
outputTokens?: number;
inputTokenDetails?: { cacheReadTokens?: number };
outputTokenDetails?: { reasoningTokens?: number };
@@ -673,6 +672,13 @@ export class StreamAgentChatJob {
providerMetadata?: Record<string, Record<string, unknown> | undefined>;
};
modelConfig: AiModelConfig;
+ streamUsage: {
+ inputTokens: number;
+ outputTokens: number;
+ inputCredits: number;
+ outputCredits: number;
+ cacheReadTokens: number;
+ };
lastStepConversationSize: number;
totalCacheCreationTokens: number;
onUpdateUsage: (usage: {
@@ -686,47 +692,44 @@ export class StreamAgentChatJob {
onUpdateCacheCreationTokens: (tokens: number) => void;
}) {
if (part.type === 'finish-step') {
- const stepInput = part.usage?.inputTokens ?? 0;
- const stepCacheCreation = extractCacheCreationTokens(
+ const stepCacheCreationTokens = extractCacheCreationTokens(
part.providerMetadata,
);
-
- onUpdateCacheCreationTokens(totalCacheCreationTokens + stepCacheCreation);
- onUpdateConversationSize(stepInput);
- }
-
- if (part.type === 'finish') {
- const breakdown = computeCostBreakdown(modelConfig, {
- inputTokens: part.totalUsage?.inputTokens,
- outputTokens: part.totalUsage?.outputTokens,
- cachedInputTokens: part.totalUsage?.inputTokenDetails?.cacheReadTokens,
- reasoningTokens: part.totalUsage?.outputTokenDetails?.reasoningTokens,
- cacheCreationTokens: totalCacheCreationTokens,
+ const stepBreakdown = computeStepCostBreakdown(modelConfig, {
+ usage: part.usage,
+ cacheCreationTokens: stepCacheCreationTokens,
});
- const inputCredits = convertDollarsToCreditsMicro(
- breakdown.inputCostInDollars,
- );
- const outputCredits = convertDollarsToCreditsMicro(
- breakdown.outputCostInDollars,
- );
-
onUpdateUsage({
- inputTokens: breakdown.tokenCounts.totalInputTokens,
- outputTokens: part.totalUsage?.outputTokens ?? 0,
- inputCredits,
- outputCredits,
- cacheReadTokens: breakdown.tokenCounts.cachedInputTokens,
+ inputTokens:
+ streamUsage.inputTokens + stepBreakdown.tokenCounts.totalInputTokens,
+ outputTokens:
+ streamUsage.outputTokens + (part.usage?.outputTokens ?? 0),
+ inputCredits:
+ streamUsage.inputCredits +
+ convertDollarsToCreditsMicro(stepBreakdown.inputCostInDollars),
+ outputCredits:
+ streamUsage.outputCredits +
+ convertDollarsToCreditsMicro(stepBreakdown.outputCostInDollars),
+ cacheReadTokens:
+ streamUsage.cacheReadTokens +
+ stepBreakdown.tokenCounts.cachedInputTokens,
});
+ onUpdateCacheCreationTokens(
+ totalCacheCreationTokens + stepCacheCreationTokens,
+ );
+ onUpdateConversationSize(part.usage?.inputTokens ?? 0);
+ }
+ if (part.type === 'finish') {
return {
createdAt: new Date().toISOString(),
usage: {
- inputTokens: breakdown.tokenCounts.totalInputTokens,
- outputTokens: part.totalUsage?.outputTokens ?? 0,
- cachedInputTokens: breakdown.tokenCounts.cachedInputTokens,
- inputCredits: toDisplayCredits(inputCredits),
- outputCredits: toDisplayCredits(outputCredits),
+ inputTokens: streamUsage.inputTokens,
+ outputTokens: streamUsage.outputTokens,
+ cachedInputTokens: streamUsage.cacheReadTokens,
+ inputCredits: toDisplayCredits(streamUsage.inputCredits),
+ outputCredits: toDisplayCredits(streamUsage.outputCredits),
conversationSize: lastStepConversationSize,
},
model: {
diff --git a/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/services/chat-execution.service.ts b/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/services/chat-execution.service.ts
index 5878d3c64bd..89ce317db8b 100644
--- a/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/services/chat-execution.service.ts
+++ b/packages/twenty-server/src/engine/metadata-modules/ai/ai-chat/services/chat-execution.service.ts
@@ -520,9 +520,9 @@ export class ChatExecutionService {
const cacheCreationTokens = extractCacheCreationTokensFromSteps(steps);
const totalTokens = (usage.inputTokens ?? 0) + (usage.outputTokens ?? 0);
- const costInDollars = this.aiBillingService.calculateCost(
+ const costInDollars = this.aiBillingService.calculateStepsCost(
registeredModel.modelId,
- { usage, cacheCreationTokens },
+ steps,
);
const creditsUsedMicro = convertDollarsToCreditsMicro(costInDollars);