fix: raise deepseek output budget for reasoning responses

frontend/src/lib/components/copilot/lib.anthropicRouting.test.ts

@@ -116,13 +116,31 @@ async function setupClients() {
 
 beforeEach(async () => {
 	await setupClients()
-})
+}, 60000)
 
 afterEach(() => {
 	vi.restoreAllMocks()
 	h.currentModel = undefined
 })
 
+describe('DeepSeek output budget', () => {
+	it.each(['deepseek-flash', 'deepseek-v4-flash', 'deepseek-v4-pro'])(
+		'sends the full reasoning and answer budget for %s',
+		async (model) => {
+			const { getCompletion, getNonStreamingMetadataCompletion, METADATA_MAX_TOKENS } =
+				await import('./lib')
+			h.currentModel = { provider: 'deepseek', model }
+
+			await getCompletion(messages, new AbortController())
+			expect(openaiCreate.mock.calls[0][0].max_tokens).toBe(393216)
+
+			openaiCreate.mockClear()
+			await getNonStreamingMetadataCompletion(messages, new AbortController())
+			expect(openaiCreate.mock.calls[0][0].max_tokens).toBe(METADATA_MAX_TOKENS)
+		}
+	)
+})
+
 describe('Anthropic Messages API routing', () => {
 	it('getNonStreamingCompletion routes Foundry Claude through the Anthropic client', async () => {
 		const { getNonStreamingCompletion } = await import('./lib')

frontend/src/lib/components/copilot/lib.ts

@@ -310,7 +310,10 @@ export async function fetchAvailableModels(
 }
 
 export function getModelMaxTokens(provider: AIProvider, model: string) {
-	if (model.includes('gpt-5')) {
+	if (/^(deepseek-flash|deepseek-v4-(flash|pro))$/.test(model.split('/').pop() ?? '')) {
+		// DeepSeek counts reasoning against the output budget as well as the answer.
+		return 393216
+	} else if (model.includes('gpt-5')) {
 		return 128000
 	} else if (
 		(provider === 'azure_openai' || provider === 'openai' || provider === 'azure_foundry') &&