No matter what, the LanguageModelSession always returns very lengthy / verbose responses. I set the maximumResponseTokens option to various small numbers but it doesn't appear to have any effect. I've even used this instructions format to keep responses between 3-8 words but it returns multiple paragraphs. Is there a way to manage LLM response length? Thanks.
I've tried the same instructions and prompt you provided with different text, and the issue doesn't happen to me: Every time I tried, the models generated a response no more than 8 words.
I built my code with Xcode 26.0 beta (17A5241e) on macOS 15.5 (24F74), and ran it on my iPhone 16 Plus + 23A5260h. The code is pretty straightforward, and so I am wondering if the test environment has any difference...
Best,
——
Ziqiao Chen
Worldwide Developer Relations.
Hi, I wanted to follow up on this. I'm now on Beta 4 for Xcode and macOS but it's still the same issue. Note that I am trying a RAG approach via Tool Calling as mentioned here:
// https://developer.apple.com/videos/play/wwdc2025/301/?time=124
// https://developer.apple.com/documentation/foundationmodels/expanding-generation-with-tool-calling
var session: LanguageModelSession
session = LanguageModelSession(
tools: [RetrievalTool(retrieval)],
instructions: instructions
)
// https://developer.apple.com/documentation/foundationmodels/generationoptions
let response = try await session.respond(
to: prompt,
options: GenerationOptions.init(maximumResponseTokens: 500)
)
tools can return a somewhat lengthy document relevant to the prompt but even though the instructions and maximumResponseTokens specifies to return a brief response, response ends up being around the same length of the tools
Thanks for your following up. If the issue is still there in the latest Beta, I’d suggest that you file an actionable feedback report against that – If you do so, please share your report ID here for folks to track.
Be sure to attach your Language model feedback, as mentioned in the link.
Best,
——
Ziqiao Chen
Worldwide Developer Relations.