> ## Documentation Index
> Fetch the complete documentation index at: https://docs.finkkle.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Finkkle AI Models: Choose the Right Model for Your Task

> Finkkle offers the Arctic model family in Spaces and GPT-5.x class models via Finkkle One — from fast lightweight completions to high-capability reasoning and voice synthesis.

<Warning>
  **Finkkle One is currently undergoing a major overhaul.** GPT-5.x class models and other Finkkle One services are unavailable until further notice. Arctic models in Spaces may also be affected. Finkkle Trends remains accessible. More details soon.
</Warning>

Finkkle provides two families of AI models: the **Arctic** series, available natively in Finkkle Spaces, and the **GPT-5.x** class models available through Finkkle One. The right model depends on your task — start with the smallest model that can reliably finish the job, and switch up when you need more capability.

## Arctic models (Finkkle Spaces)

The Arctic family is optimized for Spaces conversations. Select your model from the model picker at the bottom of the composer.

| Model          | Strengths                          | Best for                                                  |
| -------------- | ---------------------------------- | --------------------------------------------------------- |
| **Arctic 1**   | Fast, low latency                  | Quick Q\&A, short summaries, simple transformations       |
| **Arctic 1.2** | Balanced speed and quality         | Most everyday Spaces tasks                                |
| **Arctic 1.3** | Larger context, stronger reasoning | Long documents, complex analysis, multi-step planning     |
| **Arctic 1.4** | Highest Arctic capability          | Deep research, intricate reasoning, large context windows |

## GPT-5.x models (Finkkle One)

These models are available via the Finkkle One API and Studio. They require a Finkkle One account.

| Model            | Description                                                                       |
| ---------------- | --------------------------------------------------------------------------------- |
| **GPT-5.5**      | A new class of intelligence for professional and research-grade work              |
| **GPT-5.4**      | Intelligence at scale — ideal for agents, pipelines, and professional tasks       |
| **GPT-5.4-mini** | Highly efficient lightweight core engine for fast, cost-effective pipelines       |
| **Audio**        | Multi-modal acoustic model for voice synthesis, text-to-speech, and transcription |

## Choosing a model

<Accordion title="I need a fast response for a short task">
  Use **Arctic 1** or **GPT-5.4-mini**. Both are optimized for low latency and work well for classification, short summaries, and simple transformations.
</Accordion>

<Accordion title="I'm doing complex multi-step reasoning or planning">
  Use **Arctic 1.4** or **GPT-5.4**. These models handle ambiguous requirements, multi-step plans, and technical decisions best.
</Accordion>

<Accordion title="I have a very large document or codebase">
  Use **Arctic 1.3** or **Arctic 1.4** — they offer the largest context windows in the Arctic family. Make sure to set clear source boundaries when feeding large inputs.
</Accordion>

<Accordion title="I need the highest possible output quality">
  Use **GPT-5.5**. It delivers the strongest intelligence for professional-grade research, writing, and analysis.
</Accordion>

<Accordion title="I'm building an automated pipeline">
  Use **GPT-5.4-mini** for high-throughput, cost-efficient pipelines. Use **GPT-5.4** when the pipeline requires stronger reasoning or external tool use.
</Accordion>

<Accordion title="I need voice generation or transcription">
  Use the **Audio** model via the Finkkle One API. It supports text-to-speech, voice synthesis, and multi-lingual transcription.
</Accordion>

## Prompting for quality

Model selection is only part of what determines output quality. A well-structured prompt often improves results more than switching to a larger model:

* **State the outcome first** — tell the model what you need, not just the topic
* **Include the audience and format** — who is this for and what should it look like?
* **Provide only the context that matters** — focused context produces more accurate results than a large information dump
* **Ask for assumptions and uncertainties** — when accuracy matters, ask the model to flag what it is uncertain about
* **Break complex work into steps** — review each step before moving to the next

<Warning>
  Models can be wrong, incomplete, or overconfident. Always verify important facts, inspect generated code before running it, and never include secrets or personal credentials in prompts.
</Warning>