I am using the Gemini Developer API v1beta with gemini-3.8-flash on an existing free-tier project. models.get and countTokens succeed; a text-only generateContent request returned HTTP400 INVALID_ARGUMENT without usable diagnostic details on 2026-10-10 at 16:38:57 UTC. It had 67,962 counted input tokens, maxOutputTokens 8,192, candidateCount 1, thinkingLevel LOW, and no tools/cache.
The API reference/discovery defines generationConfig.responseFormat.text.mimeType as APPLICATION_JSON; the endpoint REST structured-output guide instead sends application/json. Which exact raw REST value is supported for this model? Can you identify the supported next diagnostic or prerequisite for this rejection without billing changes or private payload disclosure?
This report includes no key, headers, project/account identifiers, repository content, prompts or private provider responses. We have paused generation rather than retrying or changing models.