diff --git a/providers/cohere/chat.go b/providers/cohere/chat.go
index 14eb8bfe..fd845043 100644
--- a/providers/cohere/chat.go
+++ b/providers/cohere/chat.go
@@ -144,13 +144,9 @@ func ConvertToChatOpenai(provider base.ProviderInterface, response *CohereRespon
 		Created: common.GetTimestamp(),
 		Choices: []types.ChatCompletionChoice{choice},
 		Model:   request.Model,
-		Usage: &types.Usage{
-			PromptTokens: response.Meta.BilledUnits.InputTokens,
-		},
+		Usage:   &types.Usage{},
 	}
-
-	openaiResponse.Usage.CompletionTokens = response.Meta.BilledUnits.OutputTokens + response.Meta.Tokens.SearchUnits + response.Meta.Tokens.Classifications
-	openaiResponse.Usage.TotalTokens = openaiResponse.Usage.PromptTokens + openaiResponse.Usage.CompletionTokens
+	*openaiResponse.Usage = usageHandle(&response.Meta.BilledUnits)
 
 	usage := provider.GetUsage()
 	*usage = *openaiResponse.Usage
@@ -188,6 +184,7 @@ func (h *CohereStreamHandler) convertToOpenaiStream(cohereResponse *CohereStream
 
 	if cohereResponse.EventType == "stream-end" {
 		choice.FinishReason = types.FinishReasonStop
+		*h.Usage = usageHandle(&cohereResponse.Response.Meta.BilledUnits)
 	} else {
 		choice.Delta = types.ChatCompletionStreamChoiceDelta{
 			Role:    types.ChatMessageRoleAssistant,
@@ -206,3 +203,13 @@ func (h *CohereStreamHandler) convertToOpenaiStream(cohereResponse *CohereStream
 	responseBody, _ := json.Marshal(chatCompletion)
 	dataChan <- string(responseBody)
 }
+
+func usageHandle(token *Tokens) types.Usage {
+	usage := types.Usage{
+		PromptTokens:     token.InputTokens,
+		CompletionTokens: token.OutputTokens + token.SearchUnits + token.Classifications,
+	}
+	usage.TotalTokens = usage.PromptTokens + usage.CompletionTokens
+
+	return usage
+}