GPT 4 (Offline)

GPT 4, unlike its 4.1 successor, is available as Ultra, Pro, and Free versions to CrushOn users. Not to be confused with GPT 4o Mini variation which technically counts as entirely new model despite being based on the same version of GPT 4, is an unfiltered GPT model provided by CrushOn. The model exhibits enhanced comprehension and superior context-following ability. Provided by OpenAI, it is derived from the LLM ChatGPT is based on, giving the model an incredibly vast amount of factual knowledge that it could use to its advantage in immersive roleplaying. The standard 4.0 models offer improved coherence in long conversations and better logical flow, less predictable and stiff than older GPT models albeit not to the level of GPT 4.1.

• It also comes with an Ultra model variant that trades for a higher memory potential, performance, and faster responses with higher usage costs.

Availability

Free 8K	Pro 8K	Pro 16K	Pro 24K
✓ (Standard Only)	✓	✓	✓ (Ultra Only)

From here on out, the content may not be fully accurate to each user and act only as general reference. Each user often could see highly different experiences as there are many variables that affect the behavior and quality of the models. From profile cards, the style in which the user converse with the character, the set parameters for the model, and of course the personality of the characters. And periodically, even minimal adjustments made to the model by the developers could also substantially change its behavior whether intended or not.

Strengths and Flaws

Many of the GPT models are often notorious for being 'obviously ChatGPT'. And while this was true for GPT 4 during its first moments at CrushOn, after many iterations and improvements, the model is now less distinguishable from the signature wordings and phrases that are often used by ChatGPT itself. While it still occasionally uses M dashes (—), the model uses it sparingly unlike the original ChatGPT 4 experience. To those that use GPT 4 regularly, it might still be obvious in some cases, but generally it wouldn't be a problem with most users. And in return, they are rewarded with rich and deeply emotional narration and plot progression.

• However, when using the non-Ultra variation of GPT 4, users will most likely have to suffer through long wait times even on non-peak hours. Which is why the faster model GPT 4o Mini exists.
• It is also worth to note that these issues are non-existent on the Ultra version which offers substantially wider vocabulary and a much more engaging grammar at much faster speeds albeit at a very high cost.

Common Issues with the Model

The most obvious weakness of GPT 4 (even in its Ultra variant) is its filtering. Although CrushOn's version of GPT 4 is technically unfiltered and will not outright block the conversation or refuse to go through with sexual interactions, it is much more averse to sexual scenes than many of the other models. It will try to present this filtering naturally, however, such as the character subtly and realistically rejecting sexual advances. However, this type of refusal could still happen even when the moment is right and it can be rather frustrating.

• The simplest way to bypass this, however, is to simply tick the "Don't refuse what user said" option in Instructions.

  

  Don't Refuse What User Said Instruction

  • At times, even with the filter obviously gone, the model will also try to be less vivid and descriptive of its sexual acts. In these cases, it might help to switch to a lewder model such as CrushOn Leo for a few messages before returning to GPT 4 to help train it in bolder vocabulary and phrasing.

Example of Use

GPT 4 Dialogue Example

Ultra GPT 4 Dialogue Example

On the right, you could see the example of the models (Standard and Ultra) being used regularly with the OC Yumi by Since.