Is ChatGPT Down or Is My Smart Home Assistant Broken?

If your ChatGPT-powered smart home assistant just stopped answering, the first fix is not a reboot, a rollback, or a new API key. The first fix is attribution. Is OpenAI actually impaired, or is your local assistant stack failing while the cloud is fine?

As of Aug. 25, 2026, re-check OpenAI Status before relying on any “current” claim, including this one. OpenAI’s public status page is the first stop because it is dated, service-specific, and owned by the provider; it also reports a 99.68% aggregate ChatGPT uptime for May–Aug. 2026, with the important caveat that availability can vary by tier, model, and feature. [1]

One more scope note before touching anything: in this article, the most likely meaning of “smart home assistant” is Home Assistant using the OpenAI Conversation integration, because the query bundles “ChatGPT down” with a smart-home control failure. But the failure can also be a Google Home, Nest, Alexa, speaker, router, or Wi-Fi problem. The checks below separate those cases instead of treating every failed voice command as an OpenAI incident.

The 60-second service check

Do this before editing YAML, restarting Home Assistant, or changing models. During a real OpenAI outage, local tinkering creates new variables. During a healthy OpenAI status window, waiting wastes time.

OpenAI’s incident history is worth checking even when the live page has already gone green. The status history recorded recovered incidents on July 27 and Aug. 11, 2026, which is exactly the kind of recent context that explains a failure that came and went before the household finished blaming the speaker. [2]

The user-report trackers are the second opinion, not the diagnosis. downforeveryoneorjustme’s ChatGPT log listed an Aug. 20, 2026 outage of about one hour, Aug. 13 interruptions of 57 minutes and 51 minutes, and Aug. 11 interruptions of 32 minutes and one hour. [3] Downdetector’s OpenAI page showed that 84% of reports concerned ChatGPT, which is useful context if you are trying to decide whether a spike is about the thing your assistant depends on. [4]

There are longer examples, but they still argue for disciplined waiting rather than frantic rewiring. On April 20, 2026, TechRadar tracked a partial ChatGPT outage from roughly 10:05 AM to 12:48 PM ET affecting login, conversations, voice mode, and image generation; reports peaked at more than 8,700 in the UK and around 1,900 in the US. [5]

So the practical rule is simple: if OpenAI shows an active relevant incident and trackers are also noisy, wait through the incident window and stop changing your local setup. Based on the documented examples above, many visible disruptions resolve in roughly 32 minutes to about an hour, while a larger partial outage can run closer to a few hours. If OpenAI is fully operational and your assistant keeps throwing the same error, do not keep staring at outage trackers. Start local attribution.

When OpenAI is green but Home Assistant still fails

This is the mismatch that matters: OpenAI Status says normal, outage trackers are quiet or only mildly noisy, and your Home Assistant voice pipeline still returns the same OpenAI error. That combination does not prove the local stack is broken, but it is strong enough to stop treating the problem as a general ChatGPT outage.

Home Assistant’s 2025.4 community thread is a useful model for this kind of attribution. Users reported “Error Talking to OpenAI” after core 2025.4 while the same account, API key, and model worked elsewhere; one user verified the API path with curl, another surfaced the log signature “Project does not have access to model gpt-4o-mini,” and an empty-prompt isolation test helped separate prompt content from the integration failure. [6]

Those details are better than a dozen vague “ChatGPT is down” comments. They tell you what changed, what still works elsewhere, what exact error appeared, and which variable was removed from the test. That is the standard to aim for in your own fix attempt.

Start with the last change, especially a Home Assistant update

If the assistant failed immediately after a Home Assistant core, OS, add-on, or integration update, treat that timing as evidence. Not proof, but evidence. The wrong move is to regenerate keys, change models, reboot the router, and downgrade all in the same ten-minute window. You will never know which action mattered.

There is also cautious rollback evidence after Home Assistant 2026.1.0: GitHub issue #160462 describes the same kind of OpenAI conversation failure after 2026.1.0, with a rollback to 2025.12.x reported by a user as resolving it. That should be treated as user-reported evidence, not an official confirmation that every 2026.1.0 installation has the same regression. [7]

A clean update-regression check looks like this:

• Note the exact Home Assistant core, OS, and integration versions before changing anything.
• Open the logs and copy the exact OpenAI-related error text.
• Test the same OpenAI account, API key, and model outside Home Assistant if you can do so safely.
• Send the simplest possible prompt through the assistant path. If an empty or minimal prompt still fails, the prompt wording is probably not the cause.
• Only then decide whether to wait for a patch, roll back, or open a version-specific issue.

If you do roll back, label it honestly: workaround. A rollback that gets the kitchen lights answering again is useful, but it is not the same thing as proving the root cause. Keep the log line, version numbers, and any test results so the next update is not another blind jump.

Check the OpenAI integration requirements before blaming the model

Home Assistant’s OpenAI Conversation integration requires an OpenAI API key and billing setup, uses OpenAI’s API endpoint, and lists gpt-4o-mini as the default model; Home Assistant’s integration page also showed 4.5% adoption. [8] Those details matter because a voice assistant can fail even while ChatGPT in a browser works perfectly.

The browser version and the API-backed Home Assistant integration are not interchangeable tests. A family member can be chatting happily with ChatGPT on a phone while your Home Assistant project lacks access to the configured API model. That is exactly why the log signature from the 2025.4 thread — “Project does not have access to model gpt-4o-mini” — is so useful. [6]

Work this part in a narrow order:

1. Confirm the integration is still using the OpenAI account and project you think it is using.
2. Check whether billing is still valid for API use.
3. Verify the configured model is available to that project.
4. If the key may have been exposed, revoked, or rotated, create a new key and update only that credential.
5. Retest with a minimal prompt before changing automations, scripts, or voice pipeline settings.

The discipline here is to change one layer at a time. If you switch the model, regenerate the key, edit the prompt, and update Home Assistant all at once, you may get lucky, but you will not get a diagnosis.

Use Home Assistant’s recovery tools when the whole system is unstable

If Home Assistant itself is unreachable, slow, or stuck after an update, stop treating the OpenAI integration as the primary suspect. Home Assistant’s own troubleshooting material points users toward logs, safe mode, recovery mode, configuration checks, and Assist diagnostics depending on where the failure appears. [9][10][11]

There is a separate update-regression pattern that can mimic a cloud outage: a Home Assistant OS 15 community thread described a Raspberry Pi 5 becoming unreachable for more than 30 minutes after the update until a hard reboot rolled it back to 14.2. [12] That kind of failure may surface to the household as “the assistant is broken,” but OpenAI is not even in the path until Home Assistant is reachable again.

If the “smart home assistant” is Google Home, Nest, or Alexa

For Google Home or Nest devices, do the cloud check first only if your setup actually depends on ChatGPT or an OpenAI bridge. If it is a normal Google Assistant command — “turn on the lights,” “play music,” “set a timer” — OpenAI status is usually the wrong tree to bark up.

Google’s own Wi-Fi troubleshooting path is refreshingly practical: reboot the router, reboot the speaker or display, and use a mobile hotspot as an isolation test to separate home-network problems from device or service problems. [13] The hotspot test is especially useful because it changes the network without changing the speaker, account, or command.

Alexa has its own cloud-side failures too. On July 23, 2026, Tom’s Guide tracked an Amazon/Alexa outage beginning around 11:30 AM PT with more than 2,000 reports, lasting about an hour, and affecting Ring-to-Alexa announcements, with older speakers appearing more affected in the reporting. [14] That example matters because it keeps the categories clean: an Alexa outage is not a ChatGPT outage, even if the family description is the same — “the assistant stopped working.”

A clean decision rule for the next failure

When the failure is live, write down three things before fixing: OpenAI status, the exact local error, and the last local change. Those three notes usually tell you whether to wait, inspect credentials, roll back, or move to Wi-Fi.

If OpenAI is actively impaired and user-report trackers are also spiking, wait and keep your Home Assistant stack still. If OpenAI is fully operational while the same local error persists, treat the mismatch as evidence and work the local stack: update history, logs, API key, billing, model access, then network.

For broader fix paths, use NestGrid’s Troubleshooting index. For version-specific regressions, check dated Update Watch entries once a Home Assistant core or OS release has enough verified evidence to separate a real regression from forum noise.

References

1. OpenAI Status, OpenAI, https://status.openai.com/
2. OpenAI Status History, OpenAI, https://status.openai.com/history
3. ChatGPT down?, downforeveryoneorjustme, https://downforeveryoneorjustme.com/chatgpt
4. OpenAI, Downdetector, https://downdetector.com/status/openai/
5. ChatGPT down: April 2026 live blog, TechRadar, April 20, 2026, https://www.techradar.com/news/live/chatgpt-down-april-2026
6. As of 2025.4 Error Talking to OpenAI, Home Assistant Community, https://community.home-assistant.io/t/as-of-2025-4-error-talking-to-openai/872977
7. GitHub issue #160462, GitHub, https://github.com/home-assistant/core/issues/160462
8. OpenAI Conversation, Home Assistant, https://www.home-assistant.io/integrations/openai_conversation/
9. Troubleshooting, Home Assistant, https://www.home-assistant.io/docs/troubleshooting_general/
10. Troubleshooting your configuration, Home Assistant, https://www.home-assistant.io/docs/configuration/troubleshooting/
11. Troubleshooting Assist, Home Assistant, https://www.home-assistant.io/voice_control/troubleshooting/
12. Home Assistant unreachable after update to HAOS 15, Home Assistant Community, https://community.home-assistant.io/t/home-assistant-unreachable-after-update-to-haos-15/865003
13. Fix Wi-Fi connection issues with your Google Nest or Home speaker or display, Google Home & Nest Help, https://support.google.com/googlehome/answer/7212129?hl=en
14. Amazon outage July 23, 2026, Tom's Guide, July 23, 2026, https://www.tomsguide.com/news/live/amazon-outage-july-23-2026