Cline + Azure GPT-5: how to fix 'temperature not allowed' error?
In Cline v3.21.0 on Windows 11, I changed my Azure OpenAI deployment model from gpt-4.1 to gpt-5 (api-version 2025-04-01-preview) and Cline now fails with 'temperature not allowed'. The Azure (openai-compatible) blade seems to have no GPT-5 support. Is there a working Cline configuration for an Azure-hosted gpt-5 model?
Originally asked on GitHub: https://github.com/cline/cline/issues/5447
6 answers
Status: explained. The thread found no fix inside the Azure blade at Cline v3.21.0; the issue was closed as NOT_PLANNED, but users confirmed a working configuration through the OpenAI Compatible provider, and a later PR adds /responses support to that provider.
What happens: Azure offers two gpt-5 models. gpt-5-chat still uses the old /chat/completions URI and works in Cline, but it is text-only. gpt-5 uses the /responses URI, and Cline's Azure-compatible blade sends a temperature parameter that the endpoint rejects, producing 'temperature not allowed'. Hitting /responses directly also reports parameter mismatches: 'messages' instead of 'input', and 'max_tokens' instead of 'max_completion_tokens'.
Working configuration from the thread: In Cline, set API Provider to OpenAI Compatible. Base URL: https://XYZNAME.openai.azure.com/openai/ (opens external site) (replace XYZNAME with your Azure resource name and keep the trailing /openai/). Enter your Azure gpt-5 deployment name, and set Reasoning Effort through the Cline settings. One user confirmed requests only went through after increasing the allowed token limit.
Maintainer context: a Cline contributor said the proper fix would be to detect the gpt-5 model id in the Azure-compatible provider and set temperature to 1. Later, PR #5963 was mentioned as adding response support to the OpenAI-compatible provider, so newer Cline versions may not need this workaround.
Did this work for you?What @Erivess tried on GitHub (Worked)
Used API Provider: OpenAI Compatible with Base URL https://XYZNAME.openai.azure.com/openai/ (opens external site) (keeping the trailing /openai/), selected gpt-5, and set Reasoning Effort through settings; reported 'got it working like this'.
Source: https://github.com/cline/cline/issues/5447#issuecomment-3193801214 (opens external site)
Did this work for you?What @MESC2020 tried on GitHub (Worked)
Confirmed the gpt-5 setup from the previous comments works, but only after increasing the allowed token limit.
Source: https://github.com/cline/cline/issues/5447#issuecomment-3193900525 (opens external site)
Did this work for you?What a GitHub user tried on GitHub (Didn't work)
Changed only the model name from an Azure-deployed gpt-4.1 to gpt-5 in Cline v3.21.0 on Windows 11, keeping api-version 2025-04-01-preview; got 'temperature not allowed'.
Source: https://github.com/cline/cline/issues/5447 (opens external site)
Did this work for you?What @madsbsmail tried on GitHub (Partly worked)
Tested both Azure gpt-5 variants: gpt-5-chat works through the old /chat/completions URI with api-version 2025-01-01-preview but is text-only; gpt-5 fails with the temperature error and uses the /responses URI.
Source: https://github.com/cline/cline/issues/5447#issuecomment-3178456015 (opens external site)
Did this work for you?What @MESC2020 tried on GitHub (Didn't work)
Reported that the /responses endpoint is not supported and that calls fail with parameter mismatches: 'messages' instead of 'input', and 'max_tokens' instead of 'max_completion_tokens'.
Source: https://github.com/cline/cline/issues/5447#issuecomment-3192288378 (opens external site)
Did this work for you?