Gemini 3.7 Flash and 3.5 Flash Deprecated: Their IDs Now Route to 3.8 Flash and 3.6 Flash

Our verdict: a pinned gemini-3.7-flash or gemini-3.5-flash id no longer pins anything. Since 8 October Google answers those ids with Gemini 3.8 Flash and Gemini 3.6 Flash. Your calls still succeed, which is exactly why this is easy to miss. Treat it as a silent model swap: re-test outputs, check your bill, and change the string so your logs name the model that actually answers. Separately, the deep-research-pro-preview-12-2025 agent shuts down today, 23 October 2026.

Affected entries: Gemini 3.7 Flash and Gemini 3.5 Flash (deprecated, rerouted), Gemini 3.8 Flash and Gemini 3.6 Flash (the models now serving those calls).

What happened

• 3.7 Flash replaced by 3.8 Flash. Google's Gemini API changelog entry for 8 October 2026 says gemini-3.7-flash "is deprecated and has been replaced by gemini-3.8-flash" and tells developers to update the model string.
• 3.5 Flash replaced by 3.6 Flash. The same entry does the same for gemini-3.5-flash, replaced by gemini-3.6-flash.
• Routing, not shutdown. The deprecations page lists no shutdown date for either id. It says each endpoint "is not shutting down and automatically routes API requests" to its replacement, and that the old ids can no longer be selected in Google AI Studio.
• Deep research agent retired. The 8 October entry also deprecates deep-research-pro-preview-12-2025, with shutdown set for 23 October 2026. Google points to deep-research-preview-04-2026 for speed and efficiency and deep-research-max-preview-04-2026 for maximum comprehensiveness. Migrating means changing the agent parameter in interactions.create.

Why it matters

You lose control of which model you run. The Gemini API docs offer no way to keep calling the old model: the old ids route automatically and can no longer be selected in Google AI Studio. When 3.7 Flash went GA on 13 August, the changelog called it Google's "most intelligent workhorse model yet for coding and agents". Eight weeks later its id serves 3.8 Flash. Any eval, prompt tuning or output contract you validated on 3.7 Flash or 3.5 Flash was validated on a model you can no longer call.

Billing for routed calls is unstated. Neither the changelog nor the deprecations page says which rate applies to a call made to an old id. Google's pricing page lists Gemini 3.8 Flash and Gemini 3.6 Flash at $0.75 input and $3.75 output per 1M tokens through 31 December 2026, rising to $1.50 and $7.50 from 1 January 2027. It does not list 3.7 Flash or 3.5 Flash at all. We are not claiming a price change. We are saying nobody has published which rate applies, so read your invoice.

Per-task cost can move even if the rate does not. Our Gemini 3.8 Flash entry records Google's own caveat that 3.8 "works harder", so per-task token use runs above 3.7. A 3.7 Flash budget is not a 3.8 Flash budget.

Our ratings describe the models, not the ids. Our Gemini 3.7 Flash and Gemini 3.5 Flash entries still carry their recommended ratings. Those ratings describe the models Google released, not what the ids return today. If you want the 3.7 Flash you read about, you cannot call it through the Gemini API any more: evaluate Gemini 3.8 Flash or Gemini 3.6 Flash on their own entries instead, both rated recommended. We have not run either replacement against our own workloads.

What changes for you

1. Find every reference. Search configs, SDK calls and gateway routes for gemini-3.7-flash and gemini-3.5-flash, and change them to gemini-3.8-flash and gemini-3.6-flash.
2. Re-run evals and a cost sample on the replacement before you trust outputs or budgets set on the old model.
3. Plan for 1 January 2027. The $0.75 / $3.75 rate on both replacements ends on 31 December 2026.
4. Move deep research today. Switch deep-research-pro-preview-12-2025 to deep-research-preview-04-2026 or deep-research-max-preview-04-2026; the shutdown date is 23 October 2026.