AI Automation Suggester 1.6.2

Release date: 2026-10-02

Highlights

• GPT-6 support: OpenAI GPT-6 models such as gpt-6.1-sol and gpt-6-luna no longer fail with Unsupported parameter: 'max_tokens' is not supported with this model (#192). Model detection only recognized gpt-5, o3, and o4 names, so newer models fell back to Chat Completions with max_tokens and a temperature. GPT-5-or-later and every o-series model are now matched by version number, so future generations work without a catalog update.
  • OpenAI: these models use the Responses API with max_output_tokens and reasoning effort, as GPT-5 models already did.
  • Azure OpenAI, Custom OpenAI, Generic OpenAI: these models use Chat Completions with max_completion_tokens and no temperature.
• Safer default for unknown OpenAI models: Unrecognized model names on the direct OpenAI provider now send max_completion_tokens instead of the deprecated max_tokens. Unrecognized models on other providers are unchanged.
• hassfest compliance: aiohttp and pyyaml were removed from the manifest requirements. Home Assistant already provides both, and hassfest now rejects custom integrations that list core dependencies. The integration still uses the versions bundled with Home Assistant.

Upgrade

Update through HACS and restart Home Assistant. No config-entry or stored-history migration is required. Configured models, token budgets, and suggestion history are unchanged.

Behavior Change

• On the OpenAI provider, GPT-6-and-later and o1/o5-style model names now use the Responses API, the Reasoning Effort option, and structured output, and no longer send the configured temperature. o3 and o4-mini keep their existing catalog behavior.

Validation

• 155 automated tests pass in clean Python 3.11 and 3.12 environments, matching CI. Ruff checks pass. hassfest and HACS validation pass.
• The new regression tests fail against 1.6.1 (34 failures) and pass with this release.
• Home Assistant 2026.9.4 container smoke test: the real config flow and ai_automation_suggester.generate_suggestions service ran against a local TLS mock of api.openai.com that rejects max_tokens for GPT-5-and-later models, as the OpenAI API does.
  • 1.6.1: gpt-6.1-sol and gpt-6-luna fail with HTTP 500 and the exact sensor error from #192. gpt-5.4-mini passes.
  • 1.6.2: all three models return HTTP 200, show New Suggestions Available, and report connected.
• No live paid-provider inference was performed for this release.

Contributors

Thanks @briankavanaugh for the clear #192 report, including the control test with gpt-5.4-mini.