Gemini API Migration Checklist: 3.7 and 3.5 Flash Deprecated, Deep Research Preview Ends Oct 23

Google changed the Gemini API model lineup in two steps this month. On October 6 it made gemini-nano-banana-2.1 generally available and deprecated gemini-3.1-flash-image. On October 8 it deprecated gemini-3.7-flash and gemini-3.5-flash and set a shutdown date for the Deep Research preview agent. Anyone with those model or agent strings hard-coded in production should update them. Gemini 3.7 Flash launched in August; our launch coverage has its original positioning.

What changed

For both Flash models, Google says they are no longer selectable in Google AI Studio and that API requests are automatically routed to the replacement. That keeps calls working, but you are now running a different model than the one you tested. Pin the new id explicitly and re-run your evals.

Per-image price, old vs new Nano Banana model. Image output gets cheaper while input and text output rise. Source: Gemini API pricing.

Deep Research

The deep-research-pro-preview-12-2025 agent stops working on October 23, 2026. Google offers two successors, both released April 21, 2026. deep-research-preview-04-2026 is the faster option and suits streaming results to a UI. deep-research-max-preview-04-2026 is the most comprehensive. The change is made in the agent parameter of your interactions.create requests. Deep Research has no per-token price of its own: the pricing page says inference is billed at standard Gemini list rates, plus tool fees. Search grounding is $14 per 1,000 requests after the free monthly allotment.

Nano Banana 2.1

gemini-nano-banana-2.1 updates Nano Banana 2 (gemini-3.1-flash-image). Google cites better visual quality, prompt adherence, multi-turn character consistency and text rendering, plus wide and panoramic aspect ratios at 1K, 2K and 4K. The deprecations page also names it as the replacement for gemini-3.1-flash-image-preview (shut down Jun 25, 2026) and the three Imagen 4 models (shut down Aug 17, 2026).

Price differences

Input and text output cost three times and 2.5 times as much, while image output is half the price at 1K and 2K and about 25 percent lower at 4K. Image-heavy workloads get cheaper. Prompt-heavy, text-heavy ones may not. The old model also lists a 0.5K size ($0.045) with no 2.1 equivalent on the pricing page.

Flash pricing

The pricing page lists neither gemini-3.7-flash nor gemini-3.5-flash, so no old-versus-new comparison is possible from it. The replacements are listed: gemini-3.8-flash and gemini-3.6-flash are both $0.75 input and $3.75 output per 1M tokens through December 31, 2026, rising to $1.50 and $7.50 from January 1, 2027. Budget for the January step-up.

Checklist

1. Search code, config and infrastructure for gemini-3.7-flash, gemini-3.5-flash, gemini-3.1-flash-image and deep-research-pro-preview-12-2025.
2. Replace them with explicit new ids and re-run quality and latency tests, since silent routing has already begun for the Flash models.
3. Move Deep Research callers before October 23 and choose between the fast and max agents.
4. Re-estimate image costs using the table above.
5. Check other nearby dates: Veo 3.1 previews and gemini-omni-flash-preview shut down October 22, 2026 (replacement gemini-omni-1.1-flash).
6. Model the January 2027 Flash price increase.

Sources

• Gemini API changelog
• Gemini API deprecations
• Gemini API pricing