Add support for Deepseek V4 Pro and Deepseek V4 Flash

packages/types/src/providers/baseten.ts

@@ -83,6 +83,28 @@ export const basetenModels = {
 		description:
 			"DeepSeek's hybrid reasoning model with efficient long context scaling with GPT-5 level performance",
 	},
+	"deepseek-ai/DeepSeek-V4-Pro": {
+		maxTokens: 384_000,
+		contextWindow: 1_000_000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		inputPrice: 1.74,
+		outputPrice: 3.48,
+		cacheWritesPrice: 1.74,
+		cacheReadsPrice: 0.145,
+		description: "DeepSeek V4 Pro",
+	},
+	"deepseek-ai/DeepSeek-V4-Flash": {
+		maxTokens: 384_000,
+		contextWindow: 1_000_000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		inputPrice: 0.14,
+		outputPrice: 0.28,
+		cacheWritesPrice: 0.14,
+		cacheReadsPrice: 0.028,
+		description: "DeepSeek V4 Flash",
+	},
 	"openai/gpt-oss-120b": {
 		maxTokens: 16_384,
 		contextWindow: 128_072,

packages/types/src/providers/deepseek.ts

@@ -32,6 +32,30 @@ export const deepSeekModels = {
 		cacheReadsPrice: 0.028, // $0.028 per million tokens (cache hit) - Updated Dec 9, 2025
 		description: `DeepSeek-V3.2 (Thinking Mode) achieves performance comparable to OpenAI-o1 across math, code, and reasoning tasks. Supports Chain of Thought reasoning with up to 8K output tokens. Supports JSON output, tool calls, and chat prefix completion (beta).`,
 	},
+	"deepseek-v4-pro": {
+		maxTokens: 384_000,
+		contextWindow: 1_000_000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		preserveReasoning: true,
+		inputPrice: 1.74, // $1.74 per million tokens (cache miss)
+		outputPrice: 3.48, // $3.48 per million tokens
+		cacheWritesPrice: 1.74, // $1.74 per million tokens (cache miss)
+		cacheReadsPrice: 0.145, // $0.145 per million tokens (cache hit)
+		description: `DeepSeek V4 Pro`,
+	},
+	"deepseek-v4-flash": {
+		maxTokens: 384_000,
+		contextWindow: 1_000_000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		preserveReasoning: true,
+		inputPrice: 0.14, // $0.14 per million tokens (cache miss)
+		outputPrice: 0.28, // $0.28 per million tokens
+		cacheWritesPrice: 0.14, // $0.14 per million tokens (cache miss)
+		cacheReadsPrice: 0.028, // $0.028 per million tokens (cache hit)
+		description: `DeepSeek V4 Flash`,
+	},
 } as const satisfies Record<string, ModelInfo>
 
 // https://api-docs.deepseek.com/quick_start/parameter_settings

packages/types/src/providers/fireworks.ts

@@ -13,6 +13,8 @@ export type FireworksModelId =
 	| "accounts/fireworks/models/deepseek-v3"
 	| "accounts/fireworks/models/deepseek-v3p1"
 	| "accounts/fireworks/models/deepseek-v3p2"
+	| "accounts/fireworks/models/deepseek-v4-pro"
+	| "accounts/fireworks/models/deepseek-v4-flash"
 	| "accounts/fireworks/models/glm-4p5"
 	| "accounts/fireworks/models/glm-4p5-air"
 	| "accounts/fireworks/models/glm-4p6"
@@ -200,6 +202,28 @@ export const fireworksModels = {
 		description:
 			"DeepSeek V3.2 is the latest iteration of the V3 model family with enhanced reasoning capabilities, improved code generation, and better instruction following.",
 	},
+	"accounts/fireworks/models/deepseek-v4-pro": {
+		maxTokens: 384_000,
+		contextWindow: 1_000_000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		inputPrice: 1.74,
+		outputPrice: 3.48,
+		cacheWritesPrice: 1.74,
+		cacheReadsPrice: 0.145,
+		description: "DeepSeek V4 Pro",
+	},
+	"accounts/fireworks/models/deepseek-v4-flash": {
+		maxTokens: 384_000,
+		contextWindow: 1_000_000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		inputPrice: 0.14,
+		outputPrice: 0.28,
+		cacheWritesPrice: 0.14,
+		cacheReadsPrice: 0.028,
+		description: "DeepSeek V4 Flash",
+	},
 	"accounts/fireworks/models/glm-4p7": {
 		maxTokens: 25344,
 		contextWindow: 198000,

packages/types/src/providers/sambanova.ts

@@ -7,6 +7,7 @@ export type SambaNovaModelId =
 	| "DeepSeek-R1"
 	| "DeepSeek-V3-0324"
 	| "DeepSeek-V3.1"
+	| "DeepSeek-V4-Pro"
 	| "Llama-4-Maverick-17B-128E-Instruct"
 	| "Qwen3-32B"
 	| "gpt-oss-120b"
@@ -60,6 +61,28 @@ export const sambaNovaModels = {
 		outputPrice: 4.5,
 		description: "DeepSeek V3.1 model with 32K context window.",
 	},
+	"DeepSeek-V4-Pro": {
+		maxTokens: 384_000,
+		contextWindow: 1_000_000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		inputPrice: 1.74,
+		outputPrice: 3.48,
+		cacheWritesPrice: 1.74,
+		cacheReadsPrice: 0.145,
+		description: "DeepSeek V4 Pro",
+	},
+	"deepseek-ai/DeepSeek-V4-Flash": {
+		maxTokens: 384_000,
+		contextWindow: 1_000_000,
+		supportsImages: false,
+		supportsPromptCache: true,
+		inputPrice: 0.14,
+		outputPrice: 0.28,
+		cacheWritesPrice: 0.14,
+		cacheReadsPrice: 0.028,
+		description: "DeepSeek V4 Flash",
+	},
 	"Llama-4-Maverick-17B-128E-Instruct": {
 		maxTokens: 8192,
 		contextWindow: 131072,

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

@@ -245,6 +245,52 @@ describe("FireworksHandler", () => {
 		)
 	})
 
+	it("should return DeepSeek V4 Pro model with correct configuration", () => {
+		const testModelId: FireworksModelId = "accounts/fireworks/models/deepseek-v4-pro"
+		const handlerWithModel = new FireworksHandler({
+			apiModelId: testModelId,
+			fireworksApiKey: "test-fireworks-api-key",
+		})
+		const model = handlerWithModel.getModel()
+		expect(model.id).toBe(testModelId)
+		expect(model.info).toEqual(
+			expect.objectContaining({
+				maxTokens: 384_000,
+				contextWindow: 1_000_000,
+				supportsImages: false,
+				supportsPromptCache: true,
+				inputPrice: 1.74,
+				outputPrice: 3.48,
+				cacheWritesPrice: 1.74,
+				cacheReadsPrice: 0.145,
+				description: expect.stringContaining("DeepSeek V4 Pro"),
+			}),
+		)
+	})
+
+	it("should return DeepSeek V4 Flash model with correct configuration", () => {
+		const testModelId: FireworksModelId = "accounts/fireworks/models/deepseek-v4-flash"
+		const handlerWithModel = new FireworksHandler({
+			apiModelId: testModelId,
+			fireworksApiKey: "test-fireworks-api-key",
+		})
+		const model = handlerWithModel.getModel()
+		expect(model.id).toBe(testModelId)
+		expect(model.info).toEqual(
+			expect.objectContaining({
+				maxTokens: 384_000,
+				contextWindow: 1_000_000,
+				supportsImages: false,
+				supportsPromptCache: true,
+				inputPrice: 0.14,
+				outputPrice: 0.28,
+				cacheWritesPrice: 0.14,
+				cacheReadsPrice: 0.028,
+				description: expect.stringContaining("DeepSeek V4 Flash"),
+			}),
+		)
+	})
+
 	it("should return GLM-4.5 model with correct configuration", () => {
 		const testModelId: FireworksModelId = "accounts/fireworks/models/glm-4p5"
 		const handlerWithModel = new FireworksHandler({

src/api/providers/deepseek.ts

@@ -55,8 +55,11 @@ export class DeepSeekHandler extends OpenAiHandler {
 		const modelId = this.options.apiModelId ?? deepSeekDefaultModelId
 		const { info: modelInfo } = this.getModel()
 
-		// Check if this is a thinking-enabled model (deepseek-reasoner)
-		const isThinkingModel = modelId.includes("deepseek-reasoner")
+		// Check if this is a thinking-enabled model (deepseek-reasoner, deepseek-v4-pro, deepseek-v4-flash)
+		const isThinkingModel =
+			modelId.includes("deepseek-reasoner") ||
+			modelId.includes("deepseek-v4-pro") ||
+			modelId.includes("deepseek-v4-flash")
 
 		// Convert messages to R1 format (merges consecutive same-role messages)
 		// This is required for DeepSeek which does not support successive messages with the same role