Google Gemini Coding Tools Hit by Rate Limits, Quality Drops

Gemini-Powered Coding Tool Antigravity Faces Critical Usability Crisis

Developers relying on Antigravity, a coding assistant powered by Google's Gemini models, are reporting that the tool has become completely unusable in recent weeks. Pro-tier subscribers are encountering persistent HTTP 429 rate-limiting errors that prevent any responses from being generated — not slow responses, but zero responses — raising serious questions about Google's commitment to AI-powered coding tools.

The issues extend beyond Antigravity itself. Multiple users have noted that Gemini Flash, the underlying model powering many of Google's AI coding experiences, appears to have undergone a significant quality regression, producing buggy code that can render applications non-functional.

Key Takeaways

• Antigravity Pro users are locked out by HTTP 429 errors across all model tiers (Flash and Pro)
• The issue persists across different times of day, different versions, and different configurations
• Gemini Flash appears to have degraded significantly in coding quality over the past 2 months
• Google's Gemini CLI — a separate product from a different internal team — continues to function normally
• Developers suspect Google may be quietly deprioritizing AI coding tool resources
• For $20/month budgets, users now recommend OpenAI Codex or Claude over Google's offerings

HTTP 429 Errors Lock Out Paying Customers

The most alarming aspect of the current situation is the complete service failure for paying users. HTTP 429 is a standard 'Too Many Requests' status code, typically indicating that a user has exceeded rate limits. However, affected Antigravity users report encountering these errors even with minimal usage, suggesting the issue lies on Google's infrastructure side rather than with user behavior.

One frustrated developer documented their troubleshooting process extensively. They verified the problem persists across multiple Antigravity versions, ruling out a client-side bug. They confirmed their Pro subscription status, eliminating authentication or billing issues. And they tested at various times throughout the day, ruling out peak-hour congestion.

Perhaps most tellingly, the same developer confirmed that Google's Gemini CLI — a command-line interface for interacting with Gemini models — works perfectly fine. This is a critical data point because it suggests the rate-limiting issue is specific to Antigravity's infrastructure or API allocation, not a broader Gemini platform outage.

Meanwhile, discussions on Reddit reveal a split experience: some users report normal functionality, while others face the same complete lockout. This inconsistency points to a possible regional, account-level, or infrastructure-specific issue that Google has not publicly acknowledged.

Gemini Flash Shows Alarming Quality Regression

Beyond the availability crisis, developers are raising a separate but equally concerning alarm: Gemini Flash has gotten noticeably worse at coding tasks. Users who have been working with the model since December 2024 report a stark contrast between its earlier performance and its current output quality.

The degradation manifests in several troubling ways:

• Simple, previously reliable tasks now produce code with critical bugs
• Generated code changes can break entire applications, not just introduce minor issues
• The model appears to struggle with context retention and logical reasoning
• Quality issues appear randomly, making the tool unpredictable and untrustworthy

This pattern is familiar in the AI industry. Model providers frequently update their offerings, sometimes swapping underlying model versions without clear public communication. Google has been known to adjust model weights, quantization levels, and routing strategies behind the scenes — changes that can significantly impact real-world performance even when benchmark numbers remain stable.

For developers who built workflows around Gemini Flash's earlier capabilities, this regression represents more than an inconvenience. It undermines the fundamental value proposition of AI coding assistants: reliable, consistent augmentation of human development work. When a tool randomly introduces application-breaking bugs, it transitions from productivity booster to liability.

Is Google Quietly Deprioritizing AI Coding Tools?

The combination of infrastructure failures and model quality drops has led some developers to a provocative conclusion: Google may be shifting resources away from AI coding applications. While speculative, this theory aligns with several observable patterns in Google's recent strategic moves.

Google has been investing heavily in Gemini's multimodal capabilities, consumer-facing AI features in Search and Android, and enterprise solutions through Google Cloud. The company's AI strategy appears increasingly focused on broad horizontal applications rather than specialized developer tools. This stands in contrast to competitors like OpenAI, which recently launched Codex as a dedicated coding agent, and Anthropic, which has positioned Claude as a premium coding companion with its Claude Code product.

The broader context matters here. Google has a well-documented history of launching products with enthusiasm only to deprioritize or discontinue them later — a pattern so well-known that 'Google Graveyard' websites track hundreds of killed products. Developers who have been through this cycle before are understandably wary.

However, it is worth noting that Google recently showcased coding capabilities at Google I/O 2025, introducing improvements to Jules, its AI coding agent, and highlighting Gemini's performance on coding benchmarks. The disconnect between marketing promises and user experience is itself a red flag that suggests organizational misalignment rather than deliberate abandonment.

The Competitive Landscape Shifts Against Google

The timing of Antigravity's troubles could not be worse for Google. The AI coding assistant market is experiencing rapid consolidation around a few key players, and reliability is becoming the primary differentiator.

OpenAI's Codex, launched in 2025 as an autonomous coding agent, has quickly gained traction among developers willing to pay for consistent performance. Anthropic's Claude, particularly the Claude 4 family of models, has earned a reputation for producing high-quality, reliable code with strong reasoning capabilities. Cursor, the AI-first code editor, continues to grow its user base with tight IDE integration.

For developers evaluating where to spend their $20/month AI budget, the calculus has shifted dramatically:

• OpenAI Codex: Strong autonomous coding capabilities, reliable infrastructure, growing ecosystem
• Claude (Anthropic): Excellent code quality, superior reasoning, consistent performance
• Google Gemini tools: Competitive pricing but increasing reliability concerns
• Cursor: Best-in-class IDE integration, uses multiple model backends
• GitHub Copilot: Deep GitHub integration, powered by multiple models including GPT-4

The consensus emerging from developer communities is clear: if you can only afford one $20 subscription, Codex and Claude now rank above Google's offerings for coding-specific tasks. This represents a significant shift from just 6 months ago, when Gemini Flash's generous rate limits and competitive quality made it an attractive budget option.

What Developers Should Do Right Now

If you are currently affected by Antigravity's issues or have noticed Gemini Flash's quality decline, here are practical steps to consider.

First, document everything. Save logs, screenshots of HTTP 429 errors, and examples of degraded output. If Google does acknowledge the issue, having detailed reports will help them diagnose and fix it faster. Submit feedback through official channels — Google's AI teams do monitor developer feedback, even if response times are slow.

Second, diversify your AI coding stack. Relying on a single provider for AI-assisted development has always been risky. Consider maintaining accounts with at least 2 providers. Many developers now use Claude for complex reasoning tasks and a faster model like GPT-4o mini or Gemini Flash for simple completions.

Third, test alternatives systematically. Before committing to a new subscription, use free tiers to evaluate alternatives against your specific workflow. OpenAI offers free Codex access for limited usage, and Anthropic provides free Claude access through claude.ai.

Looking Ahead: Trust Is the Real Casualty

The most lasting damage from episodes like this is not technical — it is reputational. Developers build workflows, habits, and muscle memory around their tools. When those tools fail unpredictably, the cost extends far beyond the subscription fee.

Google faces a critical trust challenge in the AI coding space. The company has the technical talent and infrastructure to compete with anyone. But reliability, consistency, and transparent communication matter more than raw benchmark scores when developers are choosing tools for daily use.

The next few weeks will be telling. If Google addresses the Antigravity issues and stabilizes Gemini Flash's coding performance, this episode may be remembered as a temporary growing pain. If the silence continues, it will reinforce the growing narrative that Google's AI coding tools are not ready for professional reliance.

For now, developers are voting with their wallets — and increasingly, those votes are going to OpenAI and Anthropic. Google's window to reverse this trend is narrowing with each day of HTTP 429 errors and buggy code output.

Have you experienced similar issues with Antigravity or Gemini Flash? The developer community benefits from shared data points. Consider documenting your experience on forums like Reddit, Hacker News, or relevant GitHub repositories to help build a clearer picture of the scope and cause of these problems.