About 15; min
Google’s Gemini API offers the cheapest frontier-class AI models available. With Flash at $0.10 per million tokens and Pro supporting up to 2 million tokens of context, Gemini provides capabilities that no other provider matches at the same price point. Here’s the complete guide to Gemini API pricing in 2026.
Model Pricing Overview
| Model | Input (per M tokens) | Output (per M tokens) | Context | Best For |
|---|---|---|---|---|
| Gemini 2.0 Flash | $0.10 | $0.40 | 1M | Cheapest frontier model |
| Gemini 1.5 Flash | $0.075 | $0.30 | 1M | Budget tasks |
| Gemini 1.5 Pro | $1.25 | $5.00 | 2M | Complex tasks, huge context |
| Gemini 1.5 Pro (2M) | $2.50 | $10.00 | 2M | Maximum context window |
| Gemini Ultra | Custom | Custom | 1M | Maximum capability |
Gemini 2.0 Flash — The Price Breaker
Gemini 2.0 Flash is the cheapest frontier-class model from any major provider. At $0.10 per million input tokens and $0.40 per million output tokens, it undercuts GPT-4o mini ($0.15/$0.60) and Claude Haiku ($0.80/$4.00) while delivering competitive quality. The 1 million token context window dwarfs competitors — process entire books, full codebases, or hours of meeting transcripts. Flash includes multimodal input (text, images, audio, video) at no extra per-modality cost. For high-volume applications where every fraction of a cent per request matters — classification, extraction, summarization, chatbots — Flash provides frontier quality at near-open-source pricing.
Gemini 1.5 Pro — The Context Giant
Gemini 1.5 Pro at $1.25 per million input tokens delivers stronger reasoning than Flash at half the cost of GPT-4o ($2.50). The standard pricing covers up to 128K tokens of context. For requests using 128K to 2M tokens, pricing doubles to $2.50 input and $10.00 output per million. The 2 million token context window is 10x larger than Claude’s 200K and 15x larger than GPT-4o’s 128K. This enables use cases impossible on other platforms: analyze an entire codebase (500+ files), process a full-length film transcript, or ingest years of customer support tickets in a single request. For applications that need to reason over massive amounts of data in a single pass, Pro’s context window is unmatched.
Free Tier
| Model | Free Limits (AI Studio) | Rate Limits |
|---|---|---|
| Gemini 2.0 Flash | 1,500 requests/day | 15 RPM, 1M TPM |
| Gemini 1.5 Flash | 1,500 requests/day | 15 RPM, 1M TPM |
| Gemini 1.5 Pro | 50 requests/day | 2 RPM, 32K TPM |
Google AI Studio’s free tier is the most generous among major providers. 1,500 free Flash requests per day covers development, testing, and even light production workloads. Most startups can prototype and validate their AI product entirely within the free tier. The free tier includes all features — multimodal input, function calling, JSON mode, and streaming.
Additional Products
| Product | Pricing | Use Case |
|---|---|---|
| Imagen 3 (image generation) | $0.03/image (standard) | AI image generation |
| Text Embedding | Free (AI Studio) | RAG, semantic search |
| Grounding (Google Search) | $35/1K grounded requests | Real-time web info in responses |
| Context Caching | $0.025/M tokens/hour (Flash) | Reuse large contexts cheaply |
| Tuning | Free (AI Studio, limited) | Fine-tune Gemini on your data |
Cost Scenarios
High-volume chatbot (100,000 conversations/month)
Average: 400 input + 300 output tokens per conversation.
- Gemini 2.0 Flash: (40M × $0.10/M) + (30M × $0.40/M) = $4.00 + $12.00 = $16/month
- GPT-4o mini: (40M × $0.15/M) + (30M × $0.60/M) = $6.00 + $18.00 = $24/month
- Claude Haiku: (40M × $0.80/M) + (30M × $4.00/M) = $32.00 + $120.00 = $152/month
Flash saves 33% vs GPT-4o mini and 89% vs Claude Haiku at this volume.
Document analysis (10,000 long documents/month)
Average: 50,000 input tokens + 2,000 output per document.
- Gemini 1.5 Pro: (500M × $1.25/M) + (20M × $5.00/M) = $625 + $100 = $725/month
- With Context Caching (shared reference docs): ~$400/month
- GPT-4o: (500M × $2.50/M) + (20M × $10.00/M) = $1,450/month (2x more expensive)
Massive context analysis (entire codebase)
Single request: 1.5M tokens input + 5,000 output.
- Gemini 1.5 Pro (2M tier): (1.5 × $2.50) + (0.005 × $10.00) = $3.75 + $0.05 = $3.80 per analysis
- No other provider can process 1.5M tokens in a single request at any price.
Context Caching
Context Caching stores large inputs (system prompts, reference documents, code repositories) and reuses them across multiple requests. Instead of sending the same 100K-token document with every query, cache it and pay only the caching storage fee ($0.025/M tokens/hour for Flash, $0.315/M for Pro) plus the much smaller per-request input. For applications that reference the same large context repeatedly — RAG systems, document Q&A, code assistants — caching can reduce input costs by 80-90%.
Gemini vs Competitors
| Tier | Gemini | OpenAI | Anthropic |
|---|---|---|---|
| Budget | Flash: $0.10/$0.40 | 4o mini: $0.15/$0.60 | Haiku: $0.80/$4.00 |
| Standard | 1.5 Pro: $1.25/$5.00 | 4o: $2.50/$10.00 | Sonnet: $3.00/$15.00 |
| Context | 2M tokens | 128K tokens | 200K tokens |
| Free tier | 1,500 req/day (Flash) | $5 credit | $5 credit |
Gemini is the cheapest option at every tier and offers the largest context window by a wide margin. The tradeoff: many developers find GPT-4o and Claude Sonnet produce more reliable code and follow complex instructions more precisely. Gemini Flash excels at high-volume, straightforward tasks. For nuanced reasoning and coding, benchmark against GPT-4o and Sonnet on your specific use case before committing.
Gemini 2.0 Flash is the best starting point for most Gemini API applications. At $0.10/M input tokens, it provides frontier-class quality at a price that makes AI accessible for high-volume use cases. Gemini 1.5 Pro earns its premium when you need the 2M token context window or stronger reasoning on complex tasks. The free tier (1,500 requests/day) lets you build and test before spending anything. For budget-conscious AI applications, Gemini’s pricing is 2-10x cheaper than alternatives.
FAQ
Is the free tier enough for production?
At 1,500 requests per day for Flash, small applications can run entirely on the free tier. For higher volume, pay-as-you-go pricing starts immediately with no minimum commitment.
Google AI Studio vs Vertex AI — which should I use?
AI Studio for prototyping and small-scale production (simpler, generous free tier). Vertex AI for enterprise deployment (VPC integration, SLA, compliance, IAM). Same models, different wrappers. Vertex pricing may vary slightly.
Can Gemini process video?
Yes. Gemini accepts video input natively — upload a video file or provide a YouTube URL, and the model analyzes visual content, transcribes speech, and answers questions about what it sees. Video tokens count toward context limits and input pricing.
Is the 2M context window practical?
Yes, for specific use cases: entire codebase analysis, book-length document processing, multi-hour meeting transcript analysis, and video understanding. Most applications don’t need 2M tokens, but when you do, no other provider offers it.
