Choose an AI Model

Krumzi lets you pick which AI model designs for you: Claude, GPT, Gemini, or Kimi. Here's what each is good at and how to switch.

Written By Andrei Terteci

Last updated 7 days ago

Krumzi isn't tied to one AI. You can pick which model designs for you, and switch mid-conversation whenever you like. Every paid plan (and the free trial) gets the full roster and the model picker.

Where to switch models

The model picker sits in the controls next to the prompt box on the AI Designer, and in the chat panel inside the editor. Open it, pick a model, and your next message uses it. You can change models between turns without starting over. Your last-used model is remembered per brand, and you can set a default at Settings β†’ AI.

The roster

ModelProviderBest for

Claude Fable 5

Anthropic

Most powerful. Hardest briefs and highest-stakes work.

Claude Opus 5 (default)

Anthropic

Best all-round designer.

Claude Opus 4.8

Anthropic

Previous Opus. Deep reasoning for restyles.

Claude Sonnet 5

Anthropic

Strong design reasoning at balanced speed.

Claude Haiku 4.5

Anthropic

Cheapest. Quick tweaks and small edits.

GPT 5.6 Sol

OpenAI

Newest flagship GPT. Frontier reasoning and copy.

GPT 5.6 Terra

OpenAI

Balanced GPT. Smart on most briefs, fair price.

GPT 5.6 Luna

OpenAI

Cheapest GPT. Quick edits at high volume.

Kimi K3

Moonshot AI

Strong visual taste, great value. Thinking always on.

Gemini 3.1 Pro

Google

Long context. Great for big briefs.

Gemini 2.5 Pro

Google

Long context, balanced quality.

Gemini 2.5 Flash

Google

Fastest Google model. Simple tasks.

In the picker, models tagged Premium are the heavy hitters and models tagged Fast are the cheap, quick ones. Each model also shows a base credit rate next to its name so you can compare cost at a glance.

How model choice affects credits

Credits in Krumzi are usage-based: you're charged after each turn for the actual work the AI did. Two things drive the cost: which model you used and how much it had to do. A big first generation on Claude Fable 5 costs more than a one-line tweak on Haiku. Errored turns are free, and if you press Stop mid-turn you're charged half.

Thinking mode

Most models support a Think toggle for deeper reasoning. It's on by default for new sessions and it helps on complex briefs (multi-slide carousels, dense layouts, video). It's slower and uses more credits, so turn it off for quick edits. Kimi K3 always thinks, so the toggle has no effect there.

Which one should you use?

  • Not sure? Stay on Claude Opus 5. It's the default because it's the best all-rounder.

  • Quick edits ("make it bigger", "swap the color"): Claude Haiku 4.5, GPT 5.6 Luna, or Gemini 2.5 Flash.

  • Something important (client work, a campaign hero): Claude Fable 5 or GPT 5.6 Sol.

  • Huge briefs (long articles, big transcripts): Gemini 3.1 Pro.

  • Value with taste: Kimi K3.

Models interpret the same prompt differently. If one result isn't landing, try the same brief on another model before you rewrite it.

Image generation models

Separate from the chat model, the AI uses an image model when it generates artwork for your design (5 credits per image). Starter and Pro use Gemini 3.1 Flash Image; Ultra gets the upgraded Gemini 3 Pro Image. You don't pick this one, it's tied to your plan.