gemini-3.8-flash spec sheet, so the earlier “context specs not yet published” caveat is withdrawn
The model page at ai.google.dev/gemini-api/docs/models/gemini-3.8-flash is up. Key specs (source: Google’s official docs, captured 2026/9/3):
The
minimal error matches what our pre-launch tests found: the native endpoint returns 400 Thinking level is unsupported; to turn thinking off entirely use thinkingConfig: {"thinkingBudget": 0}. On the OpenAI-compatible endpoint, reasoning_effort: "minimal" does not error but is silently ignored and thinking is billed as usual, so check any code migrated from 3.6 Flash.
APIYI went live with the model on the evening of September 2. Both the default and svip groups and both the OpenAI-compatible and Gemini native endpoints are open, priced line for line with gemini-3.7-flash. With the 1M-token context now officially confirmed, the earlier advice to hold off a full cutover for long-context workloads no longer applies.
📖 Launch note with 150 paired test cases: gemini-3.8-flash Is Live, Same Price as 3.7 · Integration docs: Gemini 3.8 Flash Overview
← Back to Live Updates · 📚 Monthly archive