From 0411a01858975fc476d23b0b0e6bbb1f64eb3f38 Mon Sep 17 00:00:00 2001 From: Winston Chang Date: Tue, 29 Sep 2026 16:49:16 -0500 Subject: [PATCH] Cap DeepSeek V4 Flash max output at 262,144 on Posit AI Pass Baseten Model APIs rejects max_tokens above 262,144 for deepseek-ai/DeepSeek-V4-Flash-0731, so the 384K limit copied from DeepSeek's native API caused HTTP 400 errors. --- .../ai-config/src/model-capabilities/model-apis-helpers.ts | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/packages/ai-config/src/model-capabilities/model-apis-helpers.ts b/packages/ai-config/src/model-capabilities/model-apis-helpers.ts index 5b74502..0283acc 100644 --- a/packages/ai-config/src/model-capabilities/model-apis-helpers.ts +++ b/packages/ai-config/src/model-capabilities/model-apis-helpers.ts @@ -73,7 +73,9 @@ const MODEL_APIS_CAPABILITIES: Record> = { supportsImages: false, supportsToolResultImages: false, supportedInputMediaTypes: [], - maxOutputTokens: 384_000, + // Baseten Model APIs rejects max_tokens above 262,144 for this model, + // lower than DeepSeek's native API limit of 384K. + maxOutputTokens: 262_144, maxContextLength: 250_000, maxInputTokens: 250_000, },