Gemini 2.5 Flash-Lite
Timeline (last 90 days)
The regular expression /gemini-2\.5/i matches google/gemini-2.5-flash-lite, causing the gateway to incorrectly inject the reasoning: { effort: 'low' } parameter into that non-reasoning model.
For a model that does not reason by default, once any reasoning effort field (including low) is explicitly passed in the request, it is equivalent to asking the model to start thinking, which causes a model that should be fast to become slow or even return an empty string.
In a test with max_tokens set to 16, Gemini 2.5 Flash-Lite reported 12 reasoning tokens, with an empty visible reply.
In a test of 360 clean texts, Jev made only 1 misjudgment, gpt-5-mini made 1 misjudgment, Gemini 2.5 Flash-Lite and Gemma each made 2 misjudgments, gpt-oss-safeguard-20b made 4 misjudgments, and Llama Guard 4 made 26 misjudgments.
All Facts (4)
For a model that does not reason by default, once any reasoning effort field (including low) is explicitly passed in the request, it is equivalent to asking the model to start thinking, which causes a model that should be fast to become slow or even return an empty string.
50%UnverifiedIn a test with max_tokens set to 16, Gemini 2.5 Flash-Lite reported 12 reasoning tokens, with an empty visible reply.
50%UnverifiedIn a test of 360 clean texts, Jev made only 1 misjudgment, gpt-5-mini made 1 misjudgment, Gemini 2.5 Flash-Lite and Gemma each made 2 misjudgments, gpt-oss-safeguard-20b made 4 misjudgments, and Llama Guard 4 made 26 misjudgments.
50%UnverifiedThe regular expression /gemini-2\.5/i matches google/gemini-2.5-flash-lite, causing the gateway to incorrectly inject the reasoning: { effort: 'low' } parameter into that non-reasoning model.
50%