Reviewing Google Antigravity 2.0 and Gemini 3.5 Flash

Performance and Model Limitations

Google's new Gemini 3.5 Flash model exhibits significant price increases, reaching $1.50 per million tokens for input and $9.00 per million tokens for output. Benchmarks indicate that the model performs worse than existing frontier models like Claude 3 Opus and GPT-4. In practical testing, the Antigravity agent struggles with complex simulations, such as contact lens modeling, and produces aesthetically inferior results compared to Claude-based agents. While it demonstrates proficiency in SVG generation, it fails to provide competitive utility for real-world software engineering tasks.

UI and Integration Issues

Antigravity 2.0 suffers from a fragmented and unfinished user experience across its CLI, IDE, and standalone application. The standalone app lacks essential features like integrated terminal access and folder management, forcing users to rely on external tools. Frequent bugs include disappearing sidebars during window resizing, failure to scroll to changed code sections, and persistent authentication errors in the CLI. The interface appears to be a derivative of existing tools like Codex Desktop, yet it lacks the polish and functionality of those predecessors.

Recommended Alternatives

Given the current state of the Antigravity suite, the author suggests that users avoid the $200 subscription plan. Superior alternatives for AI-assisted coding include T3 Code, Verdant, and the GLM Coding Plan, which currently offers a 50% discount for users migrating from other platforms. Kimi is also noted as a more reliable option for development workflows.