feat: update openai models to use ai wholesaling

Author: deloreyj

apps/workers-bindings/package.json

@@ -8,7 +8,7 @@
 		"deploy": "wrangler deploy",
 		"deploy:staging": "wrangler deploy --env staging",
 		"deploy:production": "wrangler deploy --env production",
-		"eval:dev": "start-server-and-test --expect 404 eval:server http://localhost:8977 'vitest --testTimeout=60000 --config vitest.config.evals.ts --inspect=9229 --no-file-parallelism'",
+		"eval:dev": "start-server-and-test --expect 404 eval:server http://localhost:8977 'vitest --testTimeout=60000 --config vitest.config.evals.ts'",
 		"eval:server": "wrangler dev --var ENVIRONMENT:test --var DEV_DISABLE_OAUTH:true --var DEV_CLOUDFLARE_EMAIL:[email protected] --inspector-port 9230 --port 8977",
 		"eval:ci": "start-server-and-test --expect 404 eval:server http://localhost:8977 'vitest run --testTimeout=60000 --config vitest.config.evals.ts'",
 		"dev": "wrangler dev",

packages/eval-tools/src/runTask.ts

@@ -53,11 +53,6 @@ export async function runTask(
 		maxSteps: 10,
 	})
 
-	// we need to consume the fill stream, so this is empty
-	// eslint-disable-next-line no-empty
-	// for await (const _ of res.fullStream) {
-	// }
-
 	// convert into an LLM readable result so our factuality checker can validate tool calls
 	let messagesWithTools = ''
 	const toolCalls: ToolCallPart[] = []

packages/eval-tools/src/test-models.ts

@@ -20,23 +20,26 @@ type AiTextGenerationModels = Exclude<
 >
 
 function getOpenAiModel(modelName: OpenAIChatModelId) {
-	if (!env.OPENAI_API_KEY) {
-		throw new Error('No API token set!')
+	if (!env.CLOUDFLARE_ACCOUNT_ID || !env.AI_GATEWAY_ID || !env.AI_GATEWAY_TOKEN) {
+		throw new Error('No AI gateway credentials set!')
 	}
+
+	const aigateway = createAiGateway({
+		accountId: env.CLOUDFLARE_ACCOUNT_ID,
+		gateway: env.AI_GATEWAY_ID,
+		apiKey: env.AI_GATEWAY_TOKEN,
+	})
+
 	const ai = createOpenAI({
-		apiKey: env.OPENAI_API_KEY,
+		apiKey: '',
 	})
 
-	const model = ai(modelName)
+	const model = aigateway([ai(modelName)])
 
 	return { modelName, model, ai }
 }
 
 function getAnthropicModel(modelName: AnthropicMessagesModelId) {
-	if (!env.CLOUDFLARE_ACCOUNT_ID || !env.AI_GATEWAY_ID || !env.AI_GATEWAY_TOKEN) {
-		throw new Error('No AI gateway credentials set!')
-	}
-
 	const aigateway = createAiGateway({
 		accountId: env.CLOUDFLARE_ACCOUNT_ID,
 		gateway: env.AI_GATEWAY_ID,
@@ -90,6 +93,4 @@ export const eachModel = describe.each([
 	//getWorkersAiModel("@cf/meta/llama-3.3-70b-instruct-fp8-fast")
 	// Currently llama 4 is having issues with tool calling
 	//getWorkersAiModel("@cf/meta/llama-4-scout-17b-16e-instruct")
-
-	// TODO: add Claude, Gemini, new OpenAI models via AI gateway
 ])