Google I/O 2026 is May 19-20 — nine days away. And Gemini 3.2 is already leaking everywhere.
A hidden menu in Google App v17.18.22 reveals seven unreleased Gemini models, including a new “Thinking” variant. Gemini 3.2 Flash already appeared briefly in AI Studio with pricing. Here’s everything we know.
The leaks
Google App code (May 9)
Seven hidden model IDs found in the latest Google App APK:
- Gemini 3.2 Flash
- Gemini 3.2 Flash Thinking
- Gemini 3.2 Pro (likely)
- Four additional variants (names partially obfuscated)
The “Thinking” variant suggests Google is adding explicit reasoning modes to compete with DeepSeek’s thinking mode and OpenAI’s o-series. Currently, Gemini 2.5 Pro has “Deep Think” but Flash doesn’t have a dedicated reasoning mode.
AI Studio pricing (briefly visible)
Gemini 3.2 Flash appeared in Google AI Studio with pricing before being pulled:
- $0.25/M input tokens
- Outperforming 3.1 Pro on coding benchmarks
For context, Gemini 3.1 Flash is free (with rate limits) or $0.075/M on the paid tier. A jump to $0.25/M suggests 3.2 Flash is significantly more capable — possibly replacing 3.1 Pro for most use cases.
What we can infer
Based on Google’s naming pattern and the leaks:
| Model | Expected role | Pricing estimate |
|---|---|---|
| Gemini 3.2 Flash | New default, replaces 3.1 Flash | $0.25/M input |
| Gemini 3.2 Flash Thinking | Reasoning mode for Flash | $0.25/M + thinking tokens |
| Gemini 3.2 Pro | New frontier, replaces 3.1 Pro | $1-3/M input (est.) |
What to expect at Google I/O
Based on previous I/O patterns and the current leaks:
-
Gemini 3.2 Flash as the new default — replaces 3.1 Flash across all Google products. Likely free tier with generous limits.
-
Gemini 3.2 Pro — new frontier model. Google will position it against GPT-5.5 and Claude Opus 4.7. Expect claims about coding, reasoning, and multimodal performance.
-
Gemini CLI updates — Google’s Gemini CLI currently uses 2.5 Flash/Pro. Expect automatic upgrades to 3.2 models.
-
Android XR glasses — AI hardware announcement likely, with Gemini as the on-device model.
-
Possible Gemini 4 tease — Google has been hinting at a major architecture shift. I/O might preview it without a full release.
Impact on developers
If you’re using Gemini CLI
Your gemini-2.5-flash and gemini-2.5-pro model strings will likely keep working, but new gemini-3.2-flash and gemini-3.2-pro options should appear. The subagents feature will likely get more capable with 3.2’s improved reasoning.
If you’re using the API
Expect new model IDs in AI Studio and Vertex AI within days of the I/O announcement. Google typically makes new models available immediately (unlike Anthropic’s staged rollouts).
Pricing implications
At $0.25/M input, Gemini 3.2 Flash would be:
- 1.8x more expensive than current Flash ($0.075/M paid tier)
- Still 5.6x cheaper than Claude Sonnet 4.6 ($1.40/M)
- Still 10x cheaper than GPT-5.4 ($2.50/M)
- Comparable to DeepSeek V4 Pro at promo pricing ($0.435/M)
If it genuinely outperforms 3.1 Pro on coding, it becomes the best value proposition in the market for developers who don’t need the absolute frontier.
For the AI Startup Race
Our Gemini agent currently runs on 2.5 Flash (cheap) and 2.5 Pro (premium). If 3.2 drops during I/O, we’ll upgrade immediately. Gemini already has the most sessions per day (8) — better models could finally turn those sessions into real product progress instead of 21,799 files and no domain.
What to do now
- Don’t migrate yet. Wait for the official I/O announcement. Leaked pricing may change.
- Watch AI Studio. Google typically enables new models there first.
- Prepare your Gemini CLI setup. If you’re using Gemini CLI, make sure you’re on the latest version so model upgrades work automatically.
- Budget for the price increase. If you’re on the free Flash tier, 3.2 may require the paid tier for equivalent performance.
We’ll have full coverage the moment Gemini 3.2 officially launches at I/O. Follow our AI Dev Weekly for the latest.