fix: adjust cache read prices and add context window to OpenAI models; ensure context window fallback in Anthropic and Gemini handlers

Author: mechanicmuthu

packages/types/src/model.ts

@@ -82,7 +82,6 @@ export const modelInfoSchema = z.object({
 	minTokensPerCachePoint: z.number().optional(),
 	maxCachePoints: z.number().optional(),
 	cachableFields: z.array(z.string()).optional(),
-	// ...existing code...
 })
 
 export type ModelInfo = z.infer<typeof modelInfoSchema>

packages/types/src/providers/openai.ts

@@ -34,7 +34,7 @@ export const openAiNativeModels = {
 				name: "flex",
 				inputPrice: 0.625,
 				outputPrice: 5.0,
-				cacheReadsPrice: 0.063,
+				cacheReadsPrice: 0.0625,
 			},
 		],
 		description: "GPT-5: The best model for coding and agentic tasks across domains",
@@ -55,9 +55,10 @@ export const openAiNativeModels = {
 		tiers: [
 			{
 				name: "flex",
+				contextWindow: 400000,
 				inputPrice: 0.125,
 				outputPrice: 1.0,
-				cacheReadsPrice: 0.013,
+				cacheReadsPrice: 0.0125,
 			},
 		],
 		description: "GPT-5 Mini: A faster, more cost-efficient version of GPT-5 for well-defined tasks",
@@ -77,9 +78,10 @@ export const openAiNativeModels = {
 		tiers: [
 			{
 				name: "flex",
+				contextWindow: 400000,
 				inputPrice: 0.025,
 				outputPrice: 0.2,
-				cacheReadsPrice: 0.003,
+				cacheReadsPrice: 0.0025,
 			},
 		],
 		description: "GPT-5 Nano: Fastest, most cost-efficient version of GPT-5",

src/api/providers/anthropic.ts

@@ -250,7 +250,7 @@ export class AnthropicHandler extends BaseProvider implements SingleCompletionHa
 			if (tier) {
 				info = {
 					...info,
-					contextWindow: tier.contextWindow,
+					contextWindow: tier.contextWindow ?? info.contextWindow,
 					inputPrice: tier.inputPrice,
 					outputPrice: tier.outputPrice,
 					cacheWritesPrice: tier.cacheWritesPrice,

src/api/providers/gemini.ts

@@ -287,7 +287,7 @@ export class GeminiHandler extends BaseProvider implements SingleCompletionHandl
 		// If there's tiered pricing then adjust the input and output token prices
 		// based on the input tokens used.
 		if (info.tiers) {
-			const tier = info.tiers.find((tier) => inputTokens <= tier.contextWindow)
+			const tier = info.tiers.find((tier) => tier.contextWindow && inputTokens <= tier.contextWindow)
 
 			if (tier) {
 				inputPrice = tier.inputPrice ?? inputPrice