Skip to main content
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.
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.

​
GPT-5.x models (Finkkle One)

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

​
Choosing a model

Use Arctic 1 or GPT-5.4-mini. Both are optimized for low latency and work well for classification, short summaries, and simple transformations.
Use Arctic 1.4 or GPT-5.4. These models handle ambiguous requirements, multi-step plans, and technical decisions best.
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.
Use GPT-5.5. It delivers the strongest intelligence for professional-grade research, writing, and analysis.
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.
Use the Audio model via the Finkkle One API. It supports text-to-speech, voice synthesis, and multi-lingual transcription.

​
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
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.