Hello, I’d like to ask a question about an issue I encountered in the Playground.
For example, when I query in English like “What marketing automation tools suit Japanese start-ups?”, I can use the WebSearch Tool with the Combine model without any problems.
However, when I ask in Japanese, such as “日本語でおすすめのMarketingAutomationツールを調査してください” the thinking process and tool output also switch to Japanese, which seems to trigger a “Request too large” error and halts the execution.
I suspect this might be due to insufficient context length adjustments during the Combine model’s internal thinking or when processing tool outputs. Could you please advise on any solutions, or tips for tweaking parameters or prompts to work around this?
Thanks! In my case, it’s happening 100% of the time—even when I bump max_completion_tokens to 1024 or 8192. For reference, here are an other reproducible words/phrases I’ve tested below. Let me know if that helps with repro!
山梨県の絶景スポットを5つ調査して教えてください
Even in Japanese, it doesn’t always trigger an error—there’s a risk of overflow when search results include high-character elements like tables. Below are examples of queries that don’t cause errors: