Read our announcement here

We’ve added two new Experimental AI text models into our AI Dungeon lineup. GPT-4 Turbo is our first Ultra model, only available to Legend and Mythic members. Tiefighter is our newest unlimited free model available to all players.

GPT-4 Turbo is the most capable model available in AI Dungeon. It was specifically designed with improved comprehension and higher-quality generations in mind. With up to 128k tokens of context, GPT-4 Turbo could recall almost all of your Adventure, Memory, Story Cards, and more.

Tiefighter is another open-source Llama 2 13B variant. This model is even more capable at expansive and creative roleplaying in a wide range of subjects. Alpha testers have described Tiefighter as a combination of Mixtral and MythoMax for how it focuses on plot versus character emotions.

Read our announcement here

After several weeks of testing and improvements, we’re ready to take these models out of Early Access and make them official AI Dungeon models. We’ve addressed a lot of your feedback since their debut in December, and we’ve worked with our team and providers to fix many issues, primarily around empty outputs. You’re now over 50 times less likely to get an empty output on MythoMax than when we launched! We’ll continue listening to your feedback and updating these models over time.

Mixtral will also replace ChatGPT as the model we use to give free players promotional premium AI actions. ChatGPT will still be a model option for subscribed players.

We’ve implemented a server optimization for Mixtral. This change impacts the way the model is hosted, and will have no impact on the model’s behavior or performance.

Read our announcement here

We’re adding two new AI models for early access, Mixtral and MythoMax, into our lineup for AI Dungeon. Mixtral will be available for AI Dungeon Premium subscribers, and MythoMax will be available for both Premium subscribers and free players.

Mixtral is a sparse mixture-of-experts network. It's an 8x7B model, coming in at 46.7B total parameters. Industry benchmarks rank Mixtral at or above models like Llama 70B or ChatGPT (GPT-3.5).

MythoMax is a Llama 2 13B (twice as large as the current Griffin model) and was specifically optimized for storytelling.

This update features a new fine-tune for general improvements to the overall coherence and quality of AI-generated text during gameplay.

Dragon J2 is here! We heard your feedback and have been working on some model improvements to address the cliche phrases and repetition problems.

Dragon J2 V2.1 has performed 5-10% better in blind tests against our current version. Early testing has shown decreased frequency of common phrases and overall repetition during gameplay.

You can select V2.1 under the version drop down options in your game screen settings.