Compare the continuation, not just its length
Ask whether the response answers what you just said, preserves the intended speaker and gives you something useful to respond to. A longer passage is not necessarily a better fit. A model suited to quick dialogue may not be your preference for detailed narration.
Try a short scene with the choices available to your account, keeping its greeting and your persona unchanged. A consistent starting point helps you notice differences between models instead of confusing them with changes to the setup.
Find the starting plan for these model examples
Your Change Model menu shows the options currently enabled for your account. Higher tiers include lower-tier choices, but the examples here are neither a complete catalogue nor a promise of permanent availability. A model may be retired or the selection may change.
| Access starts with | Example models |
|---|---|
| Free | Default, Zeta G4 26B, Lyra Spiced, Stheno, Squelching Fantasies 8B, SpicedQ3 A3B |
| Get A Taste | Magnum 12B, Magnum Diamond 24B |
| True Supporter | Lyra 12B V4, Codex 24B, Shimizu 24B, Qwen3 235B 2507 |
| I'm All In | SpicyXL 132B, Qwen Plus Character, Glam 5.1, Kimi K2, DeepSeek V3 and V4 Pro |
Separate the model's window from your plan allowance
The effective context must fit both your account tier and the model's capacity. A model capable of a larger window cannot, by selection alone, raise the allowance on a Free account. Check both limits when comparing memory expectations.
Parameter count measures model size, while context concerns information available for a reply. Subscription access determines which models and tools you may select. These describe different properties, so one large specification does not answer all three questions.
Try a model switch before making other adjustments
To switch, open the three-dot chat menu, then Generation Settings. Under Inference Model, choose Change Model. Follow the change with a few normal exchanges before adjusting temperature, response length or the definition; otherwise several changes can obscure which one affected the result.
Switching is a comparison to try, not a certain solution to lost details or repetition. A model can still make factual errors or overlook context. When continuity is the concern, inspect the character's instructions and relevant memories too.
Questions
Is there one model I should choose for every roleplay?
No single choice suits every scene. Test available models with the same premise and compare their dialogue, pacing, instruction-following and response time against what you prefer.
Why might my context be smaller than a model's stated capacity?
The model's technical window and your subscription's allowance are separate limits. Check the active account and conversation settings; access to the model does not automatically provide its full technical capacity.