From acf1c1c5b21b7bcedd2fc7a0beff77e104d889c1 Mon Sep 17 00:00:00 2001
From: Guilhem Lemouel <guilhemlemouel@gmail.com>
Date: Tue, 22 Sep 2026 09:06:55 +0200
Subject: [PATCH] fix: scope deepseek budget and align supported model aliases

---
 .../copilot/lib.anthropicRouting.test.ts      | 34 ++++++++++++-------
 .../src/lib/components/copilot/lib.test.ts    |  1 +
 frontend/src/lib/components/copilot/lib.ts    |  6 +++-
 .../src/lib/components/copilot/modelConfig.ts |  1 +
 4 files changed, 28 insertions(+), 14 deletions(-)

diff --git a/frontend/src/lib/components/copilot/lib.anthropicRouting.test.ts b/frontend/src/lib/components/copilot/lib.anthropicRouting.test.ts
index fca7bd93a6..c2a231dedb 100644
--- a/frontend/src/lib/components/copilot/lib.anthropicRouting.test.ts
+++ b/frontend/src/lib/components/copilot/lib.anthropicRouting.test.ts
@@ -124,21 +124,29 @@ afterEach(() => {
 })
 
 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 }
+	it.each([
+		'deepseek-flash',
+		'deepseek-v4-flash',
+		'deepseek-v4-pro',
+		'deepseek-chat',
+		'deepseek-reasoner'
+	])('sends the full reasoning and answer budget for %s', async (model) => {
+		const { getCompletion } = await import('./lib')
+		h.currentModel = { provider: 'deepseek', model }
 
-			await getCompletion(messages, new AbortController())
-			expect(openaiCreate.mock.calls[0][0].max_tokens).toBe(393216)
+		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)
-		}
-	)
+	it('preserves the metadata cap and third-party provider defaults', async () => {
+		const { getNonStreamingMetadataCompletion, getModelMaxTokens, METADATA_MAX_TOKENS } =
+			await import('./lib')
+		h.currentModel = { provider: 'deepseek', model: 'deepseek-flash' }
+		await getNonStreamingMetadataCompletion(messages, new AbortController())
+		expect(openaiCreate.mock.calls[0][0].max_tokens).toBe(METADATA_MAX_TOKENS)
+		expect(getModelMaxTokens('customai', 'deepseek-flash')).toBe(8192)
+		expect(getModelMaxTokens('azure_foundry', 'deepseek-v4-pro')).toBe(8192)
+	})
 })
 
 describe('Anthropic Messages API routing', () => {
diff --git a/frontend/src/lib/components/copilot/lib.test.ts b/frontend/src/lib/components/copilot/lib.test.ts
index 964bf6e703..ad197b6b5b 100644
--- a/frontend/src/lib/components/copilot/lib.test.ts
+++ b/frontend/src/lib/components/copilot/lib.test.ts
@@ -263,6 +263,7 @@ describe('model context windows', () => {
 		expect(getKnownModelContextWindow('gemini-3-flash')).toBe(1000000)
 		expect(getKnownModelContextWindow('gemini-2.5-flash')).toBe(1000000)
 		expect(getKnownModelContextWindow('deepseek-v4-pro')).toBe(1000000)
+		expect(getKnownModelContextWindow('deepseek-flash')).toBe(1000000)
 		expect(getKnownModelContextWindow('deepseek-chat')).toBe(1000000)
 		expect(getKnownModelContextWindow('deepseek-reasoner')).toBe(1000000)
 	})
diff --git a/frontend/src/lib/components/copilot/lib.ts b/frontend/src/lib/components/copilot/lib.ts
index 239880b780..1e4396da8b 100644
--- a/frontend/src/lib/components/copilot/lib.ts
+++ b/frontend/src/lib/components/copilot/lib.ts
@@ -15,6 +15,7 @@ import { get, type Writable } from 'svelte/store'
 import { OpenAPI, ResourceService, type Script } from '../../gen'
 import { EDIT_CONFIG, FIX_CONFIG, GEN_CONFIG } from './prompts'
 import {
+	parseModelId,
 	requiresMaxCompletionTokens,
 	usesAnthropicMessagesApi,
 	usesOpenRouterPromptCaching
@@ -310,7 +311,10 @@ export async function fetchAvailableModels(
 }
 
 export function getModelMaxTokens(provider: AIProvider, model: string) {
-	if (/^(deepseek-flash|deepseek-v4-(flash|pro))$/.test(model.split('/').pop() ?? '')) {
+	if (
+		provider === 'deepseek' &&
+		/^(deepseek-(flash|chat|reasoner)|deepseek-v4-(flash|pro))$/.test(parseModelId(model).base)
+	) {
 		// DeepSeek counts reasoning against the output budget as well as the answer.
 		return 393216
 	} else if (model.includes('gpt-5')) {
diff --git a/frontend/src/lib/components/copilot/modelConfig.ts b/frontend/src/lib/components/copilot/modelConfig.ts
index cc45a23dac..18ac02de31 100644
--- a/frontend/src/lib/components/copilot/modelConfig.ts
+++ b/frontend/src/lib/components/copilot/modelConfig.ts
@@ -100,6 +100,7 @@ const MODEL_CONTEXT_WINDOWS: [name: string, contextWindow: number][] = [
 	// deepseek-reasoner aliases were retired 2026-07-24 but can still sit in a
 	// saved selection, so they keep resolving to the window they had.
 	['deepseek-v4', 1_000_000],
+	['deepseek-flash', 1_000_000],
 	['deepseek-chat', 1_000_000],
 	['deepseek-reasoner', 1_000_000],
 	['deepseek', 128_000],