A token is the basic unit of data that Gemini models use for input and output. For text, one token is about 4 characters or 0.75 words. For other media, it represents a fixed "slice" of information, such as a patch of pixels or a fraction of a second.
Token Meanings by Media Type
All media is converted into tokens to fit within the model's 1M to 2M token context window.
- Text: Approximately 4 characters per token. Standard English text averages about 750 words per 1,000 tokens.
- Images: Small images (≤384px) count as 258 tokens. Larger images are divided into 768x768 pixel blocks, each costing 258 tokens.
- Video: Typically converted at 263 tokens per second.
- Audio: Typically converted at 32 tokens per second.
- Reasoning: Models like Gemini 3.1 Progenerate internal "thinking tokens" during complex tasks, which are billed as output tokens
Token Pricing (Per 1 Million Tokens)
Pricing is pay-as-you-go, with different rates for input (data read by the model) and output (data generated by the model). Rates often double if the total context exceeds 200,000 tokens.
| Gemini Model Tier | Input Price (≤200k) | Output Price (≤200k) | Input Price (>200k) | Output Price (>200k) |
|---|---|---|---|---|
| 3.1 Pro (Preview) | $2.00 | $12.00 | $4.00 | $18.00 |
| $1.25 | $10.00 | $2.50 | $15.00 | |
| 2.5 Flash | $0.30 | $2.50 | N/A* | N/A* |
| 2.5 Flash-Lite | $0.10 | $0.40 | N/A* | N/A* |
*Flash models typically have flat pricing regardless of context length up to their limit.
Cost Reduction Features
- Context Caching: Reusing large datasets (e.g., long PDFs) can reduce input costs by 90%, with rates as low as $0.01 to $0.20 per 1M tokens.
- Batch API: Submitting non-urgent tasks for asynchronous processing provides a 50% discount on standard paid rates.
- Free Tier: Available through Google AI Studio for development, typically capped at 1,500 requests per day.
