# Gemini 2.5 Flash

> Google's fastest. 1M context.

Human version: https://kymaapi.com/models/gemini-2.5-flash
Live JSON: `GET https://kymaapi.com/v1/models` (no auth required)

## Facts

- **Model ID**: `gemini-2.5-flash` — pass this as `model` in the request body
- **Creator**: Google
- **Released**: 2025-03-20
- **Context window**: 1M tokens
- **Max output**: 8K tokens per response — a hard ceiling, not a default
- **Price**: $0.2025 in / $0.81 out per 1M, cached input 10%
- **Capabilities**: tools, vision, reasoning, caching
- **Aliases**: `fast`, `long-context` resolve to this model

## Call it

```bash
curl https://kymaapi.com/v1/chat/completions \
  -H "Authorization: Bearer $KYMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gemini-2.5-flash", "messages": [{"role": "user", "content": "Hello"}]}'
```

## Positioning

Gemini 2.5 Flash is Google’s fastest large-context model, built for high-throughput applications that process up to one million tokens. Reach for it when you need rapid multimodal ingestion, tool calling, or reasoning at a balanced cost.

## About

The model accepts text, image, audio, and video inputs, returning text outputs. It supports structured outputs, tool use, and reasoning, making it suitable for agentic workflows and standard chat applications.

On Kyma, requests route through automatic failover paths and return exact billing in usage.cost. You can target it directly or use the fast and long-context aliases. Prompt caching applies a 90% discount to repeated prefixes, and the API remains fully OpenAI-compatible at https://kymaapi.com/v1.

Output generation caps at 8,192 tokens per request. In live traffic, it maintains a median throughput of 140 tokens per second and a 97.9% success rate across 90 million served tokens.

## Use cases

- **Multimodal Document Analysis** — Ingests text, images, audio, and video to extract structured data or answer questions.
- **High-Throughput Chat Agents** — Handles rapid conversational turns with tool calling and reasoning at low latency.
- **Long-Context Summarization** — Processes up to one million input tokens to condense lengthy transcripts or logs.
- **Real-Time Media Processing** — Accepts audio and video inputs for fast transcription and content tagging.

## Not ideal for

Do not use this model for tasks requiring outputs longer than 8,192 tokens, such as generating full-length books or extensive codebases in a single pass.

## Pick something else when

- You need deeper reasoning for complex logic or math → `deepseek-r1`
- You need lower input costs for massive batch jobs → `deepseek-v4-flash`

## See also

- All models: https://kymaapi.com/models.md
- Pricing: https://kymaapi.com/pricing.md
- Other models by Google: https://kymaapi.com/models?q=Google