Fixes to all three LLM providers, prompted by two retired default models, and a --field option that lets make:crud express rules containing commas.
| Provider | Old default | Shut down | New default |
|---|---|---|---|
| OpenAI | gpt-3.5-turbo |
Oct 23, 2026 | gpt-5.6-terra |
| Gemini | gemini-2.0-flash |
Jun 1, 2026 | gemini-3.6-flash |
The replacements are the ones each vendor lists. This affects code that passes config('lara-util-x.<provider>.default_model'), which is the documented pattern.
| Provider | Fix |
|---|---|
| Gemini | System messages are sent as systemInstruction; the API rejected a system role. jsonMode now sets responseMimeType; it previously did nothing. The API key moved from the URL to the x-goog-api-key header. |
| OpenAI | $maxTokens is sent as max_completion_tokens, since reasoning models reject max_tokens. |
| Claude | Every text block is collected rather than only the first. |
| All three | Only rate limits, timeouts and server errors are retried. A missing model names the env var to change. |
See LLM Providers for details.
--fields splits on commas, so unique:users,email and in:admin,member could not be expressed. --field takes one definition per flag and is never split:
php artisan make:crud User \
--fields="name:string:required" \
--field="email:string:required|unique:users,email,NULL,id,tenant_id,7"
This also makes the v1.5.6 where-clause fix reachable from the command line. See API CRUD Generator.
OPENAI_DEFAULT_MODEL or GEMINI_DEFAULT_MODEL to a retired model in .env, update it there.openai-php/client 0.13 and later require PHP 8.2.The suite grew from 255 to 279 tests, passing in CI across Laravel 10 to 13, and now also runs weekly.