Ultra Deepseek R1
Ultra Deepseek R1, not to be confused with its predecessor available for free users, Deepseek, is one of the premium LLM options offered by CrushOn for its subscribed users, offering superior reply speed and performance at the cost of higher coin price per message. The model offers an experience that delivers natural, emotionally insightful replies, with slightly slower responses compared to its base model due to deeper processing. Provided by a Chinese AI research lab that goes by the same name, it is derived from the model Deepseek AI model, more specifically Deepseek V3, before being repurposes once more by the CrushOn team to better suit roleplay experiences.
| Free 8K | Pro 8K | Pro 16K | Pro 24K |
|---|---|---|---|
| ✕ | ✓ | ✓ | ✓ |
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
Deepseek R1 is a more powerful version of the base Deepseek model, capable of comprehending and grasping complex situations and approach it in a much more logical and realistic manner than the base Deepseek model. It is also superb at complex scenarios, world-building, and delivery of more subtle expressions. To top it off, it is also capable of story beats and cues that may not necessarily be available in any other models, giving improvisational cues for the natural progression to the user, or helping progress stories in favorable ways to name a few. Peak performances with the model will often feel like you are truly interacting with a real-life game master, allowing you to tailor your experiences and stories to the best natural outcome.
Despite the interaction provided by the model, it is not without its own faults. The model, at times, is very reliant on user's input. What this means is that any kind of flaws from user's own input such as typos, unfamiliar formatting, and grammatical errors, could affect the quality of the model's own responses (albeit not always). This dependency on the user's input may also affect the model's stability from time to time, and could rightfully be annoying to users that do not wish to perfect their own prose each time and prefer simpler responses.
Common Issues with the Model
Two of the rather common issues people seem to come to when using Deepseek R1, ironically, is very similar to its predecessor albeit to slightly differing manner. The formatting of Deepseek R1, as stated on its flaws, would at times come from the user's own manner of response.
- When issues with formatting would occur, it is best to check for any potential flaws on the user's own side of the response. Any typos or grammatical mistake may subtly influence the model.
- If no flaws like that is found, then the best option is to manually edit or regenerate the response until the preferable formatting is acquired. By that time, the model should learn from its previous iterations and will begin to answer in a much more consistent manner.
A problem more similar to its predecessor model would be how it exaggerates certain emotional traits of the OC and at times be somewhat more stubborn in them, opting for slower character developments within stories. When it comes to personality accuracy, the problem stems from Deepseek's innate quirk of wanting to deeply express the character and make the oc you interact with much more emotional. At times, some traits like playfulness and flirtatious behavior may be exaggerated, causing a dissonance in behavior when compared to other models.
- For this type of problem, the easiest way to help solve it is to write additional traits of the OC into your own profile card. Adding simple lines such as "{{char}} behaves and speaks in a soft-spoken manner" to counter an out-of-characteristically hyper character could be one such example. But of course, the manner of added personality trait should depend on the user's preference and intention.
Example of Use
On the right, you could see the example of the model being used regularly with the OC Yumi by Since.