feat: add prompt caching support for Groq provider

- Enable supportsPromptCache flag for all Groq models
- Add cacheReadsPrice with 80% discount on cached tokens
- Override createMessage to handle Groq cache metrics from prompt_tokens_details
- Update tests to verify cache token handling
- Similar implementation to Cline PR #5697

packages/types/src/providers/groq.ts

@@ -22,90 +22,100 @@ export const groqModels = {
 		maxTokens: 8192,
 		contextWindow: 131072,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 0.05,
 		outputPrice: 0.08,
+		cacheReadsPrice: 0.01, // 80% discount on cached tokens
 		description: "Meta Llama 3.1 8B Instant model, 128K context.",
 	},
 	"llama-3.3-70b-versatile": {
 		maxTokens: 8192,
 		contextWindow: 131072,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 0.59,
 		outputPrice: 0.79,
+		cacheReadsPrice: 0.118, // 80% discount on cached tokens
 		description: "Meta Llama 3.3 70B Versatile model, 128K context.",
 	},
 	"meta-llama/llama-4-scout-17b-16e-instruct": {
 		maxTokens: 8192,
 		contextWindow: 131072,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 0.11,
 		outputPrice: 0.34,
+		cacheReadsPrice: 0.022, // 80% discount on cached tokens
 		description: "Meta Llama 4 Scout 17B Instruct model, 128K context.",
 	},
 	"meta-llama/llama-4-maverick-17b-128e-instruct": {
 		maxTokens: 8192,
 		contextWindow: 131072,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 0.2,
 		outputPrice: 0.6,
+		cacheReadsPrice: 0.04, // 80% discount on cached tokens
 		description: "Meta Llama 4 Maverick 17B Instruct model, 128K context.",
 	},
 	"mistral-saba-24b": {
 		maxTokens: 8192,
 		contextWindow: 32768,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 0.79,
 		outputPrice: 0.79,
+		cacheReadsPrice: 0.158, // 80% discount on cached tokens
 		description: "Mistral Saba 24B model, 32K context.",
 	},
 	"qwen-qwq-32b": {
 		maxTokens: 8192,
 		contextWindow: 131072,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 0.29,
 		outputPrice: 0.39,
+		cacheReadsPrice: 0.058, // 80% discount on cached tokens
 		description: "Alibaba Qwen QwQ 32B model, 128K context.",
 	},
 	"qwen/qwen3-32b": {
 		maxTokens: 8192,
 		contextWindow: 131072,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 0.29,
 		outputPrice: 0.59,
+		cacheReadsPrice: 0.058, // 80% discount on cached tokens
 		description: "Alibaba Qwen 3 32B model, 128K context.",
 	},
 	"deepseek-r1-distill-llama-70b": {
 		maxTokens: 8192,
 		contextWindow: 131072,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 0.75,
 		outputPrice: 0.99,
+		cacheReadsPrice: 0.15, // 80% discount on cached tokens
 		description: "DeepSeek R1 Distill Llama 70B model, 128K context.",
 	},
 	"moonshotai/kimi-k2-instruct": {
 		maxTokens: 16384,
 		contextWindow: 131072,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 1.0,
 		outputPrice: 3.0,
+		cacheReadsPrice: 0.2, // 80% discount on cached tokens
 		description: "Moonshot AI Kimi K2 Instruct 1T model, 128K context.",
 	},
 	"openai/gpt-oss-120b": {
 		maxTokens: 32766,
 		contextWindow: 131072,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 0.15,
 		outputPrice: 0.75,
+		cacheReadsPrice: 0.03, // 80% discount on cached tokens
 		description:
 			"GPT-OSS 120B is OpenAI's flagship open source model, built on a Mixture-of-Experts (MoE) architecture with 20 billion parameters and 128 experts.",
 	},
@@ -113,9 +123,10 @@ export const groqModels = {
 		maxTokens: 32768,
 		contextWindow: 131072,
 		supportsImages: false,
-		supportsPromptCache: false,
+		supportsPromptCache: true,
 		inputPrice: 0.1,
 		outputPrice: 0.5,
+		cacheReadsPrice: 0.02, // 80% discount on cached tokens
 		description:
 			"GPT-OSS 20B is OpenAI's flagship open source model, built on a Mixture-of-Experts (MoE) architecture with 20 billion parameters and 32 experts.",
 	},

src/api/providers/__tests__/groq.spec.ts

@@ -42,6 +42,8 @@ describe("GroqHandler", () => {
 		const model = handler.getModel()
 		expect(model.id).toBe(groqDefaultModelId)
 		expect(model.info).toEqual(groqModels[groqDefaultModelId])
+		// Verify prompt caching is enabled
+		expect(model.info.supportsPromptCache).toBe(true)
 	})
 
 	it("should return specified model when valid model is provided", () => {
@@ -50,6 +52,8 @@ describe("GroqHandler", () => {
 		const model = handlerWithModel.getModel()
 		expect(model.id).toBe(testModelId)
 		expect(model.info).toEqual(groqModels[testModelId])
+		// Verify prompt caching is enabled
+		expect(model.info.supportsPromptCache).toBe(true)
 	})
 
 	it("completePrompt method should return text from Groq API", async () => {
@@ -108,7 +112,13 @@ describe("GroqHandler", () => {
 		const firstChunk = await stream.next()
 
 		expect(firstChunk.done).toBe(false)
-		expect(firstChunk.value).toEqual({ type: "usage", inputTokens: 10, outputTokens: 20 })
+		expect(firstChunk.value).toEqual({
+			type: "usage",
+			inputTokens: 10,
+			outputTokens: 20,
+			cacheWriteTokens: 0,
+			cacheReadTokens: 0,
+		})
 	})
 
 	it("createMessage should pass correct parameters to Groq client", async () => {
@@ -221,4 +231,94 @@ describe("GroqHandler", () => {
 			undefined,
 		)
 	})
+
+	it("createMessage should handle cached tokens from Groq API", async () => {
+		const testContent = "This is test content from Groq stream"
+		const cachedTokens = 50
+
+		mockCreate.mockImplementationOnce(() => {
+			return {
+				[Symbol.asyncIterator]: () => ({
+					next: vitest
+						.fn()
+						.mockResolvedValueOnce({
+							done: false,
+							value: { choices: [{ delta: { content: testContent } }] },
+						})
+						.mockResolvedValueOnce({
+							done: false,
+							value: {
+								choices: [{ delta: {} }],
+								usage: {
+									prompt_tokens: 100,
+									completion_tokens: 20,
+									prompt_tokens_details: {
+										cached_tokens: cachedTokens,
+									},
+								},
+							},
+						})
+						.mockResolvedValueOnce({ done: true }),
+				}),
+			}
+		})
+
+		const stream = handler.createMessage("system prompt", [])
+		const chunks = []
+		for await (const chunk of stream) {
+			chunks.push(chunk)
+		}
+
+		// Should have text chunk and usage chunk
+		expect(chunks).toHaveLength(2)
+		expect(chunks[0]).toEqual({ type: "text", text: testContent })
+
+		// Usage chunk should properly handle cached tokens
+		expect(chunks[1]).toEqual({
+			type: "usage",
+			inputTokens: 50, // 100 total - 50 cached = 50 non-cached
+			outputTokens: 20,
+			cacheWriteTokens: 0, // Groq doesn't track cache writes
+			cacheReadTokens: 50,
+		})
+	})
+
+	it("createMessage should handle missing cache information gracefully", async () => {
+		mockCreate.mockImplementationOnce(() => {
+			return {
+				[Symbol.asyncIterator]: () => ({
+					next: vitest
+						.fn()
+						.mockResolvedValueOnce({
+							done: false,
+							value: {
+								choices: [{ delta: {} }],
+								usage: {
+									prompt_tokens: 100,
+									completion_tokens: 20,
+									// No prompt_tokens_details
+								},
+							},
+						})
+						.mockResolvedValueOnce({ done: true }),
+				}),
+			}
+		})
+
+		const stream = handler.createMessage("system prompt", [])
+		const chunks = []
+		for await (const chunk of stream) {
+			chunks.push(chunk)
+		}
+
+		// Should handle missing cache information gracefully
+		expect(chunks).toHaveLength(1)
+		expect(chunks[0]).toEqual({
+			type: "usage",
+			inputTokens: 100, // No cached tokens, so all are non-cached
+			outputTokens: 20,
+			cacheWriteTokens: 0,
+			cacheReadTokens: 0, // Default to 0 when not provided
+		})
+	})
 })

src/api/providers/groq.ts

@@ -1,6 +1,10 @@
+import { Anthropic } from "@anthropic-ai/sdk"
+
 import { type GroqModelId, groqDefaultModelId, groqModels } from "@roo-code/types"
 
 import type { ApiHandlerOptions } from "../../shared/api"
+import type { ApiHandlerCreateMessageMetadata } from "../index"
+import { ApiStream } from "../transform/stream"
 
 import { BaseOpenAiCompatibleProvider } from "./base-openai-compatible-provider"
 
@@ -16,4 +20,42 @@ export class GroqHandler extends BaseOpenAiCompatibleProvider<GroqModelId> {
 			defaultTemperature: 0.5,
 		})
 	}
+
+	// Override to handle Groq's usage metrics, including caching
+	override async *createMessage(
+		systemPrompt: string,
+		messages: Anthropic.Messages.MessageParam[],
+		metadata?: ApiHandlerCreateMessageMetadata,
+	): ApiStream {
+		const stream = await this.createStream(systemPrompt, messages, metadata)
+
+		for await (const chunk of stream) {
+			const delta = chunk.choices[0]?.delta
+
+			if (delta?.content) {
+				yield {
+					type: "text",
+					text: delta.content,
+				}
+			}
+
+			if (chunk.usage) {
+				// Groq includes cached token information in prompt_tokens_details
+				const promptTokens = chunk.usage.prompt_tokens || 0
+				const completionTokens = chunk.usage.completion_tokens || 0
+				const cachedTokens = (chunk.usage as any).prompt_tokens_details?.cached_tokens || 0
+
+				// Calculate non-cached input tokens
+				const nonCachedInputTokens = Math.max(0, promptTokens - cachedTokens)
+
+				yield {
+					type: "usage",
+					inputTokens: nonCachedInputTokens,
+					outputTokens: completionTokens,
+					cacheWriteTokens: 0, // Groq doesn't track cache writes
+					cacheReadTokens: cachedTokens,
+				}
+			}
+		}
+	}
 }