[Feat] Add deepseek-v4-flash-vision-exp to Deepseek AI

packages/types/src/__tests__/deepseek-v4-pro.test.ts

@@ -35,6 +35,15 @@ describe("DeepSeek V4 Pro 0813 provider catalogs", () => {
 		})
 	})
 
+	it("marks deepseek-v4-flash-vision-exp as a vision-capable model", () => {
+		const model = deepSeekModels["deepseek-v4-flash-vision-exp"]
+		expect(model).toBeDefined()
+		expect(model.supportsImages).toBe(true)
+		expect(model.supportsPromptCache).toBe(true)
+		expect(model.contextWindow).toBeGreaterThanOrEqual(1_000_000)
+		expect(model.supportsReasoningEffort).toEqual(["disable", "low", "high", "max"])
+	})
+
 	// Self-hosted providers retain separate IDs for the preview weights and 0813 checkpoint.
 	it.each([
 		["Fireworks AI", fireworksModels["accounts/fireworks/models/deepseek-v4-pro"]],

packages/types/src/providers/deepseek.ts

@@ -40,6 +40,22 @@ export const deepSeekModels = {
 		cacheReadsPrice: 0.044,
 		description: `DeepSeek-V4-Pro-0813 is DeepSeek's strongest V4 model for reasoning, coding, long-context, and agentic workloads. It supports thinking and non-thinking modes, JSON output, tool calls, chat prefix completion (beta), and FIM completion (beta) in non-thinking mode.`,
 	},
+	"deepseek-v4-flash-vision-exp": {
+		displayName: "DeepSeek V4 Flash Vision Exp",
+		maxTokens: 384_000,
+		contextWindow: 1_000_000,
+		supportsImages: true,
+		supportsPromptCache: true,
+		supportsReasoningEffort: ["disable", "low", "high", "max"], // Updated 2026-08-13
+		preserveReasoning: true,
+		reasoningEffort: "high",
+		inputPrice: 0, // the inputs are priced as cache read/write, so `inputPrice` should be 0
+		// Static estimates use peak rates; off-peak rates are 50% lower.
+		outputPrice: 1.32,
+		cacheWritesPrice: 0.44,
+		cacheReadsPrice: 0.014,
+		description: `DeepSeek-V4-Flash-Vision-Exp is DeepSeek's experimental multimodal V4 Flash model with image understanding. It supports thinking and non-thinking modes, JSON output, tool calls, chat prefix completion (beta), and image inputs.`,
+	},
 } as const satisfies Record<string, ModelInfo>
 
 // https://api-docs.deepseek.com/quick_start/parameter_settings

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

@@ -249,6 +249,18 @@ describe("DeepSeekHandler", () => {
 			expect((model.info as ModelInfo).reasoningEffort).toBe("high")
 		})
 
+		it("should return vision model info for deepseek-v4-flash-vision-exp", () => {
+			const handlerWithVision = new DeepSeekHandler({
+				...mockOptions,
+				apiModelId: "deepseek-v4-flash-vision-exp",
+			})
+			const model = handlerWithVision.getModel()
+			expect(model.id).toBe("deepseek-v4-flash-vision-exp")
+			expect(model.info.supportsImages).toBe(true)
+			expect(model.info.supportsPromptCache).toBe(true)
+			expect((model.info as ModelInfo).preserveReasoning).toBe(true)
+		})
+
 		it("should return provided model ID with default model info if model does not exist", () => {
 			const handlerWithInvalidModel = new DeepSeekHandler({
 				...mockOptions,
@@ -412,6 +424,35 @@ describe("DeepSeekHandler", () => {
 			const reasoningChunks = chunks.filter((chunk) => chunk.type === "reasoning")
 			expect(reasoningChunks).toEqual([{ type: "reasoning", text: "primary thought" }])
 		})
+
+		it("enables thinking and forwards image_url for the vision model", async () => {
+			const visionHandler = new DeepSeekHandler({
+				...mockOptions,
+				apiModelId: "deepseek-v4-flash-vision-exp",
+			})
+			const withImage: Anthropic.Messages.MessageParam[] = [
+				{
+					role: "user",
+					content: [
+						{ type: "text", text: "Describe:" },
+						{ type: "image", source: { type: "base64", media_type: "image/png", data: "abc" } },
+					],
+				},
+			]
+
+			await collectStream(visionHandler.createMessage(systemPrompt, withImage))
+
+			const callArgs = mockCreate.mock.calls[0][0]
+			expect(callArgs.thinking).toEqual({ type: "enabled" })
+			expect(callArgs.messages).toEqual(
+				expect.arrayContaining([
+					expect.objectContaining({
+						role: "user",
+						content: expect.arrayContaining([expect.objectContaining({ type: "image_url" })]),
+					}),
+				]),
+			)
+		})
 	})
 
 	describe("processUsageMetrics", () => {
@@ -684,6 +725,36 @@ describe("DeepSeekHandler", () => {
 					rawReasoningEffort: "max",
 					mappedReasoningEffort: "max",
 				},
+				{
+					modelId: "deepseek-v4-flash-vision-exp",
+					rawReasoningEffort: "disable",
+					mappedReasoningEffort: undefined,
+				},
+				{
+					modelId: "deepseek-v4-flash-vision-exp",
+					rawReasoningEffort: "low",
+					mappedReasoningEffort: "low",
+				},
+				{
+					modelId: "deepseek-v4-flash-vision-exp",
+					rawReasoningEffort: "medium",
+					mappedReasoningEffort: "high",
+				},
+				{
+					modelId: "deepseek-v4-flash-vision-exp",
+					rawReasoningEffort: "high",
+					mappedReasoningEffort: "high",
+				},
+				{
+					modelId: "deepseek-v4-flash-vision-exp",
+					rawReasoningEffort: "xhigh",
+					mappedReasoningEffort: "high",
+				},
+				{
+					modelId: "deepseek-v4-flash-vision-exp",
+					rawReasoningEffort: "max",
+					mappedReasoningEffort: "max",
+				},
 			]
 
 			for (const { modelId, rawReasoningEffort, mappedReasoningEffort } of mappings) {

src/api/providers/deepseek.ts

@@ -28,7 +28,7 @@ type DeepSeekChatCompletionParams = Omit<OpenAI.Chat.ChatCompletionCreateParamsS
 	reasoning_effort?: "low" | "high" | "max"
 }
 
-const deepSeekV4ThinkingModels = new Set(["deepseek-v4-flash", "deepseek-v4-pro"])
+const deepSeekV4ThinkingModels = new Set(["deepseek-v4-flash", "deepseek-v4-pro", "deepseek-v4-flash-vision-exp"])
 const supportsDeepSeekThinkingToggle = (modelId: string) => deepSeekV4ThinkingModels.has(modelId)
 
 // Only known V4 models and the legacy reasoner alias support DeepSeek's
@@ -51,6 +51,7 @@ export const normalizeDeepSeekReasoningEffort = (
 	switch (modelId) {
 		case "deepseek-v4-flash":
 		case "deepseek-v4-pro":
+		case "deepseek-v4-flash-vision-exp":
 			switch (reasoningEffort) {
 				case "low":
 					return "low"