Give us Gemini 3.7; your Gemini 3.8 is overrated and useless for simple scripting tasks - it spends three minutes hallucinating just to find a minor bug.
The Argon model is obviously going to be useless because it will consume excessive tokens; there is no need to sacrifice Gemini 3.7 for it. You are losing valuable users and high-quality post-training data.
I’ve heard that 3.8 is a product of RSI, which is why it loses its intelligence when dealing with real-world code.
You know the internal details better than anyone. Either release a new, stable 3.9 or keep Gemini 3.7 Flash running; nobody wants to deal with the heavy usage demands of Argon.
Hi @Jenil_Sheth,
Welcome to the Forum and thank you for sharing your feedback.
To address your points:
- Gemini 3.7 Flash: The Models page currently lists Gemini 3.8 Flash as the Gemini model available on all plans, while Gemini 3.7 Flash is now available only on Enterprise plans.
- Gemini 4 Argon: Argon is designed for complex, demanding tasks (see the launch post), and in Antigravity it’s currently available on Google AI Ultra and eligible Enterprise plans. It’s an additional option, and Gemini 3.8 Flash remains available for everyday tasks.
- Tips for simple tasks: In the model selector, try a lower thinking level (i.e.Low) for quick scripting and point the agent to the specific file and change you want so it doesn’t explore unrelated code.
To help us investigate the hallucination and latency issues you mentioned with Gemini 3.8 Flash, could you share a few example prompts or a short description of the tasks where it struggled? Please avoid posting any private or sensitive code on the public forum. If your issue involves proprietary code, we recommend using the in-app feedback option directly on those conversations so the details go straight to the team.
“fix this ” was basically enough for “this” to become fixed a minute later, and no other frontier model compared against that level of to-the-point-edness as the gemini flash 3.7