# Models changed just now while working

**URL:** <https://discuss.ai.google.dev/t/models-changed-just-now-while-working/145419>\
**Category:** Google Antigravity\
**Tags:** models\
**Created:** [May 19, 2026, 6:40pm UTC](https://discuss.ai.google.dev/t/models-changed-just-now-while-working/145419 "2026-05-19T18:40:23Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Roberto\_Garcia](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/roberto_garcia/32/104770_2.png) [@Roberto\_Garcia](https://discuss.ai.google.dev/u/Roberto_Garcia)\
**Post date:** [May 19, 2026, 6:40pm UTC](https://discuss.ai.google.dev/t/models-changed-just-now-while-working/145419/1 "2026-05-19T18:40:23Z")

</div>

Anybody notice that?  
So, i was forced to use a heavy token cosumer that eat 60% of my quota in just ONE planing to make an html row of cards!!!  
Where´s Gemini Flash? one hour ago i was working as allways.

 ![image](https://d3qe71uytubmmx.cloudfront.net/original/3X/0/f/0f2082b42d00d205445441b8a8f093263bde707a.png)

---

<div class="post-metadata">

**Author:** ![Roberto\_Garcia](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/roberto_garcia/32/104770_2.png) [@Roberto\_Garcia](https://discuss.ai.google.dev/u/Roberto_Garcia)\
**Post date:** [May 19, 2026, 7:09pm UTC](https://discuss.ai.google.dev/t/models-changed-just-now-while-working/145419/2 "2026-05-19T19:09:19Z")

</div>

Ah… well, if anybody asking same, here´s why including some TIPS:):

========================================================================  
ANTIGRAVITY / GEMINI API UPDATES SUMMARY (MAY 2026)

1. WHAT HAPPENED TO THE 5-HOUR QUOTA?

* * *

- The Culprit: The prompt accidentally ran on “Gemini 3.5 Flash (High)”.
- The Cause: In Antigravity 2.0, the (High) tag activates the Full  
Autonomous Agent Mode.
- Behind the Scenes: Instead of just returning a text response, the IDE  
spawned multiple sub-agents in parallel (planning, code generation,  
and live UI simulation to check the Bootstrap layout).
- The Impact: This massive parallel processing triggered a huge  
computation spike, draining 60% of the global 5-hour rolling quota  
in a single click.

1. THE NEW MODEL STRUCTURE IN ANTIGRAVITY 2.0

* * *

Google changed the naming to an infrastructure-based tier system:

- Gemini 3.5 Flash (Low) → LOWEST COST (SAFE)  
Best for: Daily coding, simple Bootstrap layouts, PHP/ACF logic.

- Gemini 3.1 Pro (Low) → MODERATE COST  
Best for: Longer refactoring without agent overhead.

- GPT-OSS 120B (Medium) → MEDIUM COST  
Best for: Flat cost per prompt, predictable generation.

- Claude Sonnet/Opus 4.6 (Thinking) → HIGH COST  
Best for: Complex logic (generates heavy hidden thinking tokens).

- Gemini 3.5 Pro / Flash (High) → HIGHEST COST (DANGEROUS)  
Best for: Autonomous multi-file execution and background loops.

1. GOOGLE AI STUDIO API VS. ANTIGRAVITY SUBSCRIPTION

* * *

- Independent Pools: Using your API Key bypasses the Antigravity 5-hour  
compute limits entirely.
- Pricing Difference:
  - “gemini-3.1-flash-lite” is still the cheapest ($0.075 / 1M tokens)  
for bulk tasks.
  - The new “gemini-3.5-flash” ($1.50 / 1M tokens) is much smarter and  
faster for active development, but it costs more than Lite.

1. KEY TAKEAWAYS TO PROTECT YOUR WORKFLOW

* * *

- Always keep “Gemini 3.5 Flash (Low)” selected as your default chat model.
- Keep using inline Tab Completions (they remain unlimited and free).
- 
# Upgrading to Antigravity 2.0 is safe: it will automatically migrate your local settings.json, keybindings, and custom AI skills.