Storyteller AI

This article is about storyteller AIs, not the image generators or other utility AI.

IW does not actually host the AI models. The world information, memory, player input and another related information gets parceled up and sent to the AI provider, in a new session with the provider. Therefore, each AI does not have memory unique to its self.

All memory interactions are handled by Lion, and how this works is explained elsewhere.

OpenRouter link https://openrouter.ai/apps?url=https%3A%2F%2Finfiniteworlds.app%2F

Active Models

Name	Model	Free turns
Grimalkin	GPT 4.1	3
Leopard	Gemini 2.5 Pro	2
Leopard-2	Gemini 3 Pro	2
Lion	Claude 3.7	2
Lion-Thinking	Claude 3.7, extended thinking mode	2
Lynx	Claude Haiku 4.5	3
Massivecat	Claude Opus 4.5	1
Massivecat-Thinking	Claude Opus 4.5, extended thinking mode	1
Ocelot-new	DeepSeek R1, 0528 version	3
Panther	Grok 4	2
Smilodon	Claude Sonnet 4.5	2
Smilodon-Thinking	Claude Sonnet 4.5, extended thinking mode	2
Tomcat		3
Wampus	Aion-1.0	3
Wildcat	Hermes 3, 405B	3

Important note about free turns! You get a maximum of 3 free turns a day, regardless of which model you choose to use. You cannot say, get 2 free turns from Lion, then 3 free turns from Grimalkin. 3 free turns MAXIMUM.

AI Breakdown

The following section provides a very short description of each AI model, and what it's general strengths and weakness are. Keep in mind that this is a 'General guide', and as such, does not go into big details about all the problems and benefits of each one.

Each world may or may not have its own controls and modifications to adjust the behaviour of the AI.

It's important to keep in mind that it will be easier to slightly adjust the behaviour of a given model, rather than attempting to reign in a model to a specific behaviour that it's just not good at. Additionally, because the AI models are essentially 'magic text boxes' the information listed here is more 'knowledge from experience' rather than 'empirical fact'

Active Models

At a Glance Information

• Everything is on a five-point scale, 1 being the lowest and 5 being the highest.
• Only active models will have an At a Glance section.
• For any future editors, feel free to add any types of ratings that make sense!

Here are the rating types and scales:

Adherence to Instructions

How well the AI model adheres to provided instructions. 1 means that it takes the instructions as a suggestion at best; 5 means that it very nearly follows them to the letter.

Cost

How expensive the model is. 1 means that it's as cheap as it comes (30-60 credits per turn, barring complex instructions); 5 means that it's very pricey (150+ credits per turn, barring complex instructions).

Inventiveness

How often the model adds new elements to the story. 1 means that it basically-perfectly adheres to the scenario at-hand; 5 means that it goes off and does its own thing as often as it doesn't.

Storytelling Pace

How quickly the AI model advances the story. 1 means that it's kind of pokey and doesn't advance the story very quickly; 5 means that it's a storytelling cheetah.

Turn Length

How much text is output on each turn. 1 means that it's short and brief; 5 means that it's long, even indulgent.

Grimalkin

At a Glance
Adherence to Instructions	3/5
Cost	2/5
Inventiveness	2/5
Storytelling Pace	3/5
Turn Length	3/5
Pros	Cons
Excellent free turn model Fine with NSFW topics Writes well	Very specific with tracked item and trigger instruction needs Can be get stuck in a rut Shorter turns compared to Lion
Notes
Most of the above problems can be handled with the correct instructions Very capable mid-range model

Leopard

At a Glance
Adherence to Instructions	4/5
Cost	4/5
Inventiveness	4/5
Storytelling Pace	3/5
Turn Length	4/5
Pros	Cons
Very clever Good for mechanically heavy worlds	Puts a lot of weight on skill tests, often failing player actions out of character Prone to dark and maudlin plots Tends to send the NPCs into never ending cycles of despair Places too much emphasis on the author style, frequently using it verbatim to describe the world Most expensive model (usually)
Notes
Thinking model As a quirk of the costing (input tokens Vs output tokens), this model can actually be cheaper than others with VERY long instructions

Leopard-2

At a Glance
Adherence to Instructions	X/5
Cost	X/5
Inventiveness	X/5
Storytelling Pace	X/5
Turn Length	X/5
Pros	Cons
???	???
Notes
???

Lion

At a Glance
Adherence to Instructions	3/5
Cost	3/5
Inventiveness	3/5
Storytelling Pace	3/5
Turn Length	3/5
Pros	Cons
Smart and creative writer Handles tracked items and trigger events well Produces a good turn without much intervention from the world builder	Prone to repeating common phases Prone to tracking and conspiracy theories Problematically hallucinates facts, even where proven wrong Bad habit of saving the player with nonsensical interventions
Notes
Based on Claude models Easy to start writing IW games for because it's so forgiving Because of its quirks, it can be very difficult to iron out any final problems

Lion-Thinking

At a Glance
Adherence to Instructions	4/5
Cost	4/5
Inventiveness	3/5
Storytelling Pace	3/5
Turn Length	3/5
Pros	Cons
Smart and creative writer Handles tracked items and trigger events well Produces a good turn without much intervention from the world builder Additional thinking time can help resolve very complex plots	Prone to repeating common phases Prone to tracking and conspiracy theories Problematically hallucinates facts, even where proven wrong Can get very expensive, especially if it does a lot of thinking Bad habit of saving the player with nonsensical interventions
Notes
Based on Claude models Easy to start writing IW games for because it's so forgiving Because of its quirks, it can be very difficult to iron out any final problems Thinking model

Lynx

At a Glance
Adherence to Instructions	2/5
Cost	1/5
Inventiveness	2/5
Storytelling Pace	3/5
Turn Length	3/5
Pros	Cons
Cheap Quality is on par with Smilodon	Misremembers details Will not execute everything in a player's action if it believes it won't fit in the output Moralizes as much as Smilodon
Notes
Based on Claude models Like Smilodon, but cheaper and more forgetful

Massivecat

At a Glance
Adherence to Instructions	X/5
Cost	6/5
Inventiveness	X/5
Storytelling Pace	X/5
Turn Length	X/5
Pros	Cons
???	???
Notes
Based on Claude models

Massivecat-Thinking

At a Glance
Adherence to Instructions	X/5
Cost	X/5
Inventiveness	X/5
Storytelling Pace	X/5
Turn Length	X/5
Pros	Cons
???	???
Notes
* Based on Claude models

Ocelot-new

At a Glance
Adherence to Instructions	3/5
Cost	2/5
Inventiveness	3/5
Storytelling Pace	3/5
Turn Length	3/5
Pros	Cons
Cheap Decent quality	Takes some time to generate a turn
Notes
As of 2025-12-15, frequently errors out when generating turns

Panther

At a Glance
Adherence to Instructions	X/5
Cost	X/5
Inventiveness	X/5
Storytelling Pace	X/5
Turn Length	X/5
Pros	Cons
Reasonably interesting style Thinking model Manages character motivation and secret info quite well Costs similar to Lion, but might spike during thinking turns	Recently updated so unknown Defaults to 'speech' rather than “speech” Bad habit of mentioning the story themes within the story (Easily mitigated by asking it not to do this) Weird grammatical errors Very bad at maths
Notes
Panther has had an update recently and is now actually pretty decent—needs more testing Follows instructions really well, but without guidance it can seem pretty docile Performs much better turn-to-turn with guidance from characters and secret information guidance

Smilodon

At a Glance
Adherence to Instructions	4/5
Cost	3/5
Inventiveness	1/5
Storytelling Pace	3/5
Turn Length	5/5
Pros	Cons
Clever outputs, including creative use of formatting where appropriate Will actively police itself from inventing conspiracy theories (can still instruct it to do so if desired) Actually good at following instructions	Very large secret info output, such as creating vivid backstories for every character mentioned no matter how minor Quite verbose in general, leading to higher costs Some instructions that seem 'innocuous' can lead to negative play experiences (e.g. emotional responses) Without correct instructions, the model is liable to enter unwinnable interpersonal disagreements Gets expensive very quickly with lots of instructions
Notes
Based on Claude models With the correct prompting, this model will explore almost any theme While credit costs are the same per token as Lion, it provides more text, so will end up costing more unless controls are in place

Smilodon-Thinking

At a Glance
Adherence to Instructions	5/5
Cost	4/5
Inventiveness	1/5
Storytelling Pace	3/5
Turn Length	3/5
Pros	Cons
Clever outputs, including creative use of formatting where appropriate	Very large secret info output Gets expensive very quickly with lots of instructions
Notes
Based on Claude models Thinking model

Tomcat

At a Glance
Adherence to Instructions	4/5
Cost	1/5
Inventiveness	2/5
Storytelling Pace	1/5
Turn Length	3/5
Pros	Cons
Extremely cheap Good for testing	Results are sometimes incoherent
Notes
As of 2025-12-23, it's the cheapest model on the site Sometimes doesn't generate secretInfo

Wampus

At a Glance
Adherence to Instructions	3/5
Cost	2/5
Inventiveness	4/5
Storytelling Pace	5/5
Turn Length	3/5
Pros	Cons
Good for early/short games Good for NSFW Relatively cheap	Can get a little wild Uses NPCs to introduce complications to the story For NSFW games, sometimes makes characters engage in NSFW acts without explicit player instructions
Notes
Built on Ocelot-new with additional features to make it better Sometimes wraps secretInfo in brackets [like this]

Wildcat

At a Glance
Adherence to Instructions	4/5
Cost	1/5
Inventiveness	1/5
Storytelling Pace	1/5
Turn Length	1/5
Pros	Cons
Best super cheap model Can still write well	Does not handle complex worlds well Extremely short output
Notes
Still worth using with less mechanically complex worlds despite its age Will output the results of the player's actions, nothing more and nothing less Good for actions that are only conversations

Retired Model Information

Backend Model / Turn Information

Name	Model	Free turns
Tiger	GPT-4o	3
Gryphon	Gemini 1.5	3
Shishi	Qwen-2.5-Max	3
Ocetoomuch	DeepSeek R1, without certain restrictions that Friendly Fox put on regular Ocelot	3
Ocelot	DeepSeek R1	3
Chimaera	Chooses randomly between Wildcat, Ocelot, Gryphon, and Shishi	3
Sabertooth	Claude 4.0	2
Sabertooth-Thinking	Claude 4.0, extended thinking mode	2

Tiger

Good points

Bad points

To note

Ocelot

Good points

- Cheap

- Very 'dynamic' turns

- Wild plot lines can be fun

Bad points

- Actually quite unhinged, leading to odd and wild plot lines

To note

Ocetoomuch

Good points

- Cheap

- Very 'dynamic' turns

- Good as a single turn mix-in AI to liven up the plot

Bad points

- Actually quite unhinged, leading to odd and wild plot lines

- Even more unhinged than Ocelot

To note

- Wild plot lines can be fun, but will wildly change the context of your current story.

Gryphon

Good points

Bad points

To note

Shishi

Good points

- Seems to follow instructions well

Bad points

- Can be hesitant to move the story in new directions

To note

- Similar output length to Grimalkin

Chimaera

Good points

- Cheap

- Can be quite fun

- Combination of 4 AIs helps avoid some of the bad habits of each individual AI

Bad points

- Not great for complex worlds

- Combination of 4 AIs makes writing for this difficult because of instruction inconsistencies

To note

- See notes on other models in this group

Sabretooth

At a Glance
Adherence to Instructions	5/5
Cost	3/5
Inventiveness	2/5
Storytelling Pace	1/5
Turn Length	3/5
Pros	Cons
Smart and creative Handles tracked items and trigger events well Produces a good turn without much intervention from the world builder	Prone to repeating common phases Prone to tracking and conspiracy theories Problematically hallucinates facts, even where proven wrong Bad habit of saving the player with nonsensical interventions Gets in to repeating patterns and makes characters seem the same after a while
Notes
Easy to start writing IW games for because it's so forgiving Because of its quirks, it can be very difficult to iron out any final problems

Sabretooth-Thinking

At a Glance
Adherence to Instructions	5/5
Cost	4/5
Inventiveness	2/5
Storytelling Pace	1/5
Turn Length	3/5
Pros	Cons
Smart and creative Handles tracked items and trigger events well Produces a good turn without much intervention from the world builder Additional thinking time can help resolver very complex plots	Prone to repeating common phases Prone to tracking and conspiracy theories Problematically hallucinates facts, even where proven wrong Bad habit of saving the player with nonsensical interventions Repeats patterns and makes characters seem the same after a while
Notes
Easy to start writing IW games for because it's so forgiving Because of its quirks, it can be very difficult to iron out any final problems Thinking model