After upgrading to 2026.1.0 my voice assistant responds “Error talking to OpenAI”

[zBernie]

The problem

My voice assistant is configured to use OpenAI. I restarted HA again and cycled power to the voice assistant to no avail. It was working just prior to upgrading to HA 2026.1.0. Now it just responds “Error talking to OpenAI”.

Here is the error in the log file when I say "Hey Jarvis":

What version of Home Assistant Core has the issue?

core-2026.1.0

What was the last working version of Home Assistant Core?

No response

What type of installation are you running?

Home Assistant OS

Integration causing the issue

OpenAI

Link to integration documentation on our website

No response

Diagnostics information

home-assistant_openai_conversation_2026-01-08T00-41-23.703Z.log

Example YAML snippet

Anything in the logs that might be useful for us?

Additional information

No response

[nathan-curtis]

+1 same issue.

Upgrade to 2026.1.x
Attempt to use a Voice with OpenAI
Error 400, same as Bernie*s error log.

Note for info maybe unrelated.I have an additional bug posted for what turned out to be a similar presentation in the android companion but... Maybe similar underlying contribution factors?

[sosaudio1]

+1 same condition.

Error talking to OpenAI: Error code: 400 - {'error': {'message': "Invalid schema for function 'HassStartTimer': schema must have type 'object' and not have 'oneOf'/'anyOf'/'allOf'/'enum'/'not' at the top level.", 'type': 'invalid_request_error', 'param': 'tools[4].parameters', 'code': 'invalid_function_parameters'}}

STT - "What's the forecast?"
TTS - "Error talking to OpenAI

Installation method
Home Assistant OS
Core
2026.1.0
Supervisor
2026.01.0
Operating System
16.3
Frontend
20260107.0
Companion app
2025.11.4-full (19134)

Wyoming

Conversation Agent
ChatGPT
Model
gpt-4o-mini (noted new update allows for OpenAI 5.2 - gpt-5.2.... Should we test against this model???)
Prefer Handling Commands Locally
Yes
Speech to Text
Home Assistant Cloud
Language - American English
Text to Speech
Home Assistant Cloud
Language - American English
Voice - Andrew
Streaming Wake Word Engine
openwakeword
Wake Word
Hey Jarvis

I suspect with the new change to gpt-5.2 this may have the issue pigeon holed. This happened before with the 3.5 Turbo that I used before. Could be that we need to update instances to 5.2 for testing.

[nathan-curtis]

It's an error with an intent not the model. Model should not matter. I'm in fact using 5.1mini

(Also as a side note and completely off topic 5.2 is way overkill you can get everything you want from 5.1 mini for a fraction of cost. Only use it if you have a reason.)

[t0m1o1]

Having the same issue. Turning off assist allows you to talk to it again.

[nathan-curtis]

Good to know. That supports the busted intent theory

Also completely useless for most of us unless you're just doing summarization jobs.

[t0m1o1]

Tried different models. Stripping prompts back. Unexposing everything. All to no avail

Getting the exact same error

{'error': {'message': "Invalid schema for function 'HassStartTimer': schema must have type 'object' and not have 'oneOf'/'anyOf'/'allOf'/'enum'/'not' at the top level.", 'type': 'invalid_request_error', 'param': 'tools[4].parameters', 'code': 'invalid_function_parameters'}}

[sosaudio1]

It's an error with an intent not the model. Model should not matter. I'm in fact using 5.1mini

(Also as a side note and completely off topic 5.2 is way overkill you can get everything you want from 5.1 mini for a fraction of cost. Only use it if you have a reason.)

Off-topic reply: Good to know. Any reason to move from gpt-4o-mini to gpt-5.1-mini?

[nathan-curtis]

Yes very much more capable. Test time compute. If you want to discuss further take it to the community site. Sent via the Samsung Galaxy S25+, an AT&T 5G smartphone Get Outlook for Android<https://aka.ms/AAb9ysg>

…

________________________________ From: sosaudio1 ***@***.***> Sent: Wednesday, January 7, 2026 9:09:15 PM To: home-assistant/core ***@***.***> Cc: Nathan Curtis ***@***.***>; Comment ***@***.***> Subject: Re: [home-assistant/core] After upgrading to 2026.1.0 my voice assistant responds “Error talking to OpenAI” (Issue #160462) [https://avatars.githubusercontent.com/u/33795846?s=20&v=4]sosaudio1 left a comment (home-assistant/core#160462)<#160462 (comment)> It's an error with an intent not the model. Model should not matter. I'm in fact using 5.1mini (Also as a side note and completely off topic 5.2 is way overkill you can get everything you want from 5.1 mini for a fraction of cost. Only use it if you have a reason.) Off-topic reply: Good to know. Any reason to move from gpt-4o-mini to gpt-5.1-mini? — Reply to this email directly, view it on GitHub<#160462 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AF5LLQVBDOZTL6GPLR3KOS34FXC5XAVCNFSM6AAAAACRAGNCM6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTOMRRGY4TENZWHE>. You are receiving this because you commented.Message ID: ***@***.***>

[rwb-bowman]

Same issue, rolled back to previous core and is working, skipped the update pending fix.

[sosaudio1]

Same issue, rolled back to previous core and is working, skipped the update pending fix.

Dumb question

How do I roll back correctly?

[sosaudio1]

Same issue, rolled back to previous core and is working, skipped the update pending fix.

Dumb question

How do I roll back correctly?

I figured it out. I also found out rolling back supervisor isn't as easy.... I have not done that yet

[slettmayer]

Having the same issue with OpenAI conversation after upgrade to 2026.1.0 also.
I am using gpt-4.1-mini as model.
An AI Task in a script with OpenAI integration still works. Maybe that helps to narrow it down.

[MilanDasek]

Same issue.

[cmug81]

Same issue after upgrade. Using OpenAI gpt-4o-mini-2024-07-18..