Google schedules the deprecation of Gemini 2.5 Flash for October 16, 2026, and this change necessitates that developers migrate their production workflows to the Gemini 3.5 Flash or Gemini 3.1 Flash-Lite models to avoid service interruptions. The current pricing for Gemini 2.5 Flash sits at $0.30 per 1 million input tokens and $2.50 per 1 million output tokens. This model remains an efficient option for everyday tasks. Developers must plan their migration strategies before the October deadline to ensure continuous service for their applications.
The 3.5 Flash upgrade path
Gemini 3.5 Flash launched on May 19, 2026, at Google I/O. This model provides a 1 million token context window and a 64K output window. The pricing for Gemini 3.5 Flash is $1.50 per 1 million input tokens and $9.00 per 1 million output tokens. This model outperforms Gemini 3.1 Pro on coding and agentic benchmarks. It achieves a 76.2% score on Terminal-Bench 2.1, an 83.6% score on MCP Atlas, and an 84.2% score on CharXiv Reasoning.
The Gemini 3.5 Flash model is 25% cheaper than Gemini 3.1 Pro. It also works faster than other frontier models. The model provides native multimodality for text, image, audio, and video. Use this model if your application requires higher coding performance than previous generations. You should audit your current token usage before the October deadline arrives.
Low latency via Flash-Lite
Gemini 2.5 Flash-Lite is a lightweight model that Google released on July 22, 2025. The model costs $0.10 per 1 million input tokens and $0.40 per 1 million output tokens. It provides a context window of 1,048,576 tokens. On benchmarks, Gemini 2.5 Flash-Lite scores 6.7 for Intelligence and 72.4 for MMLU Pro. It also achieves a 47.4 score for GPQA and 35.3 for Math.
Gemini 2.5 Flash-Lite provides a context window of 1,048,576 tokens. The model costs $0.10 per 1 million input tokens and $0.40 per 1 million output tokens. It matches or exceeds performance on logic and math benchmarks. Developers select this model for high-volume, low-complexity tasks. It provides 3x cost savings for summarization pipelines at enterprise scale. This model is 2x faster in throughput than Gemini 2.5 Flash for classification tasks.
The thinking token expense
Gemini 2.5 Pro and Gemini 3.1 Pro are thinking models. These models reason before they respond. This reasoning generates thinking tokens. Google bills these thinking tokens at the standard output token rate. On complex reasoning tasks, thinking tokens account for 70% to 85% of the total output bill.
Developers report unexpected cost spikes when they migrate to thinking models because the reasoning tokens count as output tokens. A prompt that generates 500 visible output tokens might consume 3,000 thinking tokens. The invoice shows 3,500 output tokens at the output rate. For Gemini 2.5 Pro, the output rate is $10.00 per 1 million tokens. For Gemini 3.1 Pro, the output rate is $12.00 per 1 million tokens for context up to 200,000 tokens and $18.00 per 1 million tokens for context over 200,000 tokens.
Will the migration to 3.5 Flash resolve the unexpected cost spikes reported by developers using thinking models?
Enterprise bundling and Antigravity
Google bundles Antigravity, its autonomous coding agent, into Gemini Enterprise Standard and Plus subscriptions. This tool is available at no additional cost for eligible users. Developers can activate Antigravity in Visual Studio Code, JetBrains, Visual Studio, and Zed. It also works with the Antigravity 2.0 desktop app and CLI.
Gemini Enterprise subscriptions include enterprise-grade spend management. Administrators can set project-level budget caps in the billing console. The platform provides pooled token quotas that flex across teams. Users can also set monthly spend caps with overage billing. The system includes centralized usage dashboards to track token consumption and API calls by team.
Security controls for Antigravity include configurable workspace sandboxing. Administrators control browser and MCP server access. The platform includes a single-toggle for comprehensive audit logging of prompts and responses. All activity runs within the Google Cloud security boundary.
Competitive API pricing comparison
The Gemini API maintains a significant price advantage for low-cost tasks. Gemini 2.5 Flash-Lite costs $0.10 per 1 million input tokens. This is 300x cheaper than the $30.00 per 1 million input tokens for GPT-4. Gemini 2.5 Flash is also much cheaper than GPT-5.5. Gemini 2.5 Flash input costs $0.30 per 1 million tokens, while GPT-5.5 input costs $5.00 per 1 million tokens. Gemini 2.5 Flash output costs $2.50 per 1 million tokens, while GPT-5.5 output costs $30.00 per 1 million tokens.
| Model | Input Price (per 1M) | Output Price (per 1M) |
|---|---|---|
| Gemini 2.5 Flash-Lite | $0.10 | $0.40 |
| Gemini 3.5 Flash | $1.50 | $9.00 |
| Gemini 3.1 Flash-Lite | $0.25 | $1.50 |
| Gemini 3.1 Pro (<=200k) | $2.00 | $12.00 |
| Gemini 3.8 Flash (thru 2026) | $0.75 | $3.75 |
The competition remains fierce in the high-end tier. Gemini 3.1 Pro costs $2.00 per 1 million input tokens for context up to 200,000 tokens. Anthropic Claude Fable 5.1 costs $10.00 per 1 million input tokens. OpenAI GPT-6 Astra also costs $10.00 per 1 million input tokens.
Vertex AI vs AI Studio
Google offers different pricing for Google AI Studio and Vertex AI. AI Studio is for developers and testing. Vertex AI is for enterprise production environments. Vertex AI pricing runs 10% to 20% higher than AI Studio for the same model. This premium covers enterprise-grade data handling and SLA guarantees. It also includes integration with Google Cloud IAM and VPC controls.
AI Studio is the cheapest way to test Gemini models. It offers a free tier for Flash and Flash-Lite models. This free tier includes 1,500 requests per day for Flash and 50 requests per day for 2.5 Pro. AI Studio usage is free in all available regions. Vertex AI is the correct choice for production deployments that require high rate limits and security.
Cost optimization strategies
Developers can reduce Gemini API costs through several methods. The Batch API offers a 50% cost reduction for non-urgent workloads. This API provides a 24-hour processing SLA. For example, Gemini 2.5 Flash-Lite drops to $0.05 per 1 million input tokens and $0.20 per 1 million output tokens with the Batch API.
Context caching reduces repeated-prompt costs by up to 90%. For Gemini 3.1 Pro, cached input costs $0.20 per 1 million tokens for context up to 200,000 tokens. For Gemini 3.5 Flash, cached input costs $0.15 per 1 million tokens. Users pay for cache storage at a rate of $1.00 to $4.50 per million tokens per hour.
Routing requests to the cheapest capable model is an effective strategy. Use Gemini 2.5 Flash-Lite for simple classification and Gemini 3.5 Flash for agentic tasks. This approach prevents unnecessary spending on premium models. Use the following table to plan your model selection.
| Task Type | Recommended Model | Reason |
|---|---|---|
| Simple classification | Gemini 2.5 Flash-Lite | Lowest unit price |
| High-throughput batch | Gemini 3.5 Flash-Lite | Fast and cheap |
| General conversation | Gemini 3.6 Flash | Current workhorse |
| Complex reasoning | Gemini 3.1 Pro Preview | Quality requirements |
| Long context | Gemini 2.5 Pro | Cost-effective long context |
