Questsmith just changed something that most AI RPG platforms would not dare touch. The free plan now lets you pick which AI model runs your adventure. Not a watered-down version, not a different tier, the actual model powering your story. You get three options: ChatGPT 5.4, Qwen 3.7 Plus, and Sakana AI Fugu. No upgrade needed. Just open the platform and choose.
On paper that sounds like a feature checkbox. In practice, it changes the character of every campaign you run. Different models write differently. They handle combat differently. They pace stories differently. They remember differently, or seem to, within a session.
I ran the same campaign opening across all three to see what the actual differences look like. Here is what I found.
Why the Model Choice Matters More Than You Think
Most people assume AI models are mostly interchangeable for creative writing. Give them the same prompt and you get roughly the same output, just worded slightly differently.
That is true for short, isolated prompts. It stops being true once you are running a campaign with a specific tone, a cast of established characters, and a story that needs to go somewhere meaningful.
The model you pick determines how the narrative voice sounds. Whether combat reads as tactical or cinematic. Whether NPCs feel like distinct personalities or variations on the same template. Whether the story generates tension or just describes events.
None of that is visible in a single test prompt. It emerges across an hour of play. That is what I tested.
The Test Setup
I used the same scenario across all three models: a fantasy setting, a solo rogue character with high Stealth and low Combat stats, a city infiltration mission in the opening act. No backstory pre-loaded, just the genre, the character build, and the starting situation.
I ran each for approximately 45 minutes, covering the same sequence of events: initial scene-setting, one social encounter, one stealth attempt that succeeded, one stealth attempt that failed, and a confrontation at the end of the act.
The D20 mechanics, persistent memory system, and companion AI stayed the same across all three, those are platform-level features that do not change with the model. What changed was the writing itself.
ChatGPT 5.4 — The Most Versatile Storyteller
ChatGPT 5.4 is the most immediately impressive. The prose is polished. Scene descriptions have atmosphere. Dialogue feels like it was written rather than generated.
The social encounter stood out. My rogue tried to bluff a warehouse guard with a half-constructed story about being a delivery worker. ChatGPT 5.4 did not just accept or reject the bluff. The guard asked a follow-up question that my story had no answer for. I had to improvise, and the model tracked whether my improvisation was consistent with what I had already said.
That kind of follow-through, where NPCs remember and respond to the specifics of what you said, not just whether you passed or failed a check, is what separates a story from a game session. ChatGPT 5.4 did this consistently throughout the run.
The combat writing leaned cinematic. When the stealth attempt failed and I ended up in a confrontation, the description of the fight read like a chase sequence. Atmospheric, fast-paced, specific about physical positioning. If you want your campaigns to read like fiction, this is the model that produces that.
Best for: Players who prioritize narrative quality, want NPC interactions that feel genuinely dynamic, and are running genres where atmosphere matters, horror, mystery, historical fiction.
Qwen 3.7 Plus — The Most Mechanically Sharp
Qwen 3.7 Plus writes differently. The prose is clean but drier. Less literary. What it trades in atmosphere it makes up for in mechanical precision.
The stealth sequence showed the difference clearly. When my rogue moved through the warehouse, Qwen 3.7 Plus tracked spatial details with more consistency than the other models. It knew which direction I had come from. It remembered that I had noted a guard patrol pattern two scenes earlier and used that detail when I attempted the infiltration. The world felt like it had a layout rather than just a setting.
The failed stealth attempt was also handled well. Qwen 3.7 Plus generated a consequence that felt connected to the specific way the failure happened, I had been rushing because of a time constraint established earlier in the session, and the failure reflected that. Not just 'you got caught' but 'you got caught because you cut a corner you would not have cut otherwise.'
NPC dialogue was functional rather than distinctive. The guard in the social encounter spoke like a guard, not like a person. That is a limitation compared to ChatGPT 5.4. But if the campaign is mechanics-heavy, if you care about tactical choices, stat-driven outcomes, and world consistency, Qwen 3.7 Plus rewards that kind of play more reliably.
Best for: Players running DnD-style campaigns where mechanical outcomes matter, stats should feel meaningful, and world consistency across a long session is the priority.
Sakana AI Fugu — The Most Surprising
Sakana AI Fugu is the least familiar name on the list. Built in Japan by Sakana AI, it is not a model most Western players would have encountered before Questsmith added it.
The writing has a different rhythm. Sentences are often shorter. Scene transitions are abrupt in a way that reads more like a film cut than a novel paragraph. At first it felt sparse. Thirty minutes in, I understood what it was doing.
Fugu generates tension differently. It does not build atmosphere through description, it builds it through what it leaves out. The warehouse scene had fewer adjectives than the same scene in ChatGPT 5.4, but the silence and the gaps in information created a different kind of unease. Things felt uncertain in a way that made decisions feel heavier.
The social encounter was the most interesting test. My rogue's bluff failed — same situation as the other models, same dice outcome. Fugu's response was a guard who did not immediately react with aggression. He went quiet. He looked at me for a moment. Then he asked a question that had nothing to do with my cover story. It was a completely different kind of pressure than the other models generated.
Fugu is not the right model for every campaign. If you want lush fantasy description or mechanical precision, the other two do those things better. But for horror, psychological tension, or any story where what is unsaid matters as much as what is said, Fugu produces something the other models do not.
Best for: Horror campaigns, psychological narratives, mystery genres, and players who want a storytelling style that feels genuinely different from the standard Western AI writing voice.
Side by Side — The Same Moment, Three Ways
To make the comparison concrete, here is how each model handled the same moment: my rogue entering a torch-lit corridor after successfully picking a lock.
ChatGPT 5.4
The final pin settles into place with a quiet click. You push the door open just enough to slip inside, where flickering torchlight reveals an ancient corridor lined with weathered stone and faded carvings. The air is cool, carrying the scent of dust and forgotten history. Each careful step echoes softly as the passage stretches into darkness, inviting both curiosity and caution.
Qwen 3.7 Plus
The lock yields after one final adjustment, and the heavy door opens with barely a whisper. Warm torchlight spills across uneven stone, exposing worn footprints, scattered debris, and walls marked by centuries of neglect. The silence is almost tangible, broken only by the faint crackle of distant flames as uncertainty lingers with every step forward.
Sakana AI Fugu
The lock surrenders with unexpected ease. Beyond the doorway, the corridor seems less like a passage and more like a memory waiting to be disturbed. Torchlight bends strangely against the walls, shadows lingering longer than they should. Every footstep feels borrowed, as though the silence itself is listening for a mistake.
Three completely different experiences from the same scenario. None of them wrong. All of them useful for different kinds of play.
Which Model Should You Pick
The honest answer is: it depends on what kind of campaign you are running.
- Running a story-first campaign — fantasy, historical fiction, anything where the writing should feel like a novel: ChatGPT 5.4.
- Running a mechanics-first campaign — DnD-style, stat-heavy, tactical choices with real consequences: Qwen 3.7 Plus.
- Running horror, mystery, or psychological narrative — anything where atmosphere and dread matter more than description: Sakana AI Fugu.
There is also nothing stopping you from switching between campaigns. Run a fantasy campaign on ChatGPT 5.4 and a horror campaign on Fugu. The persistent memory system is platform-level, it works the same regardless of which model you pick. Your 500 tracked memories per adventure carry forward whether you switch models or not.
What This Actually Means for the Free Plan
The free tier giving access to three production-level AI models is not a minor update. Most platforms gate model selection behind paid tiers specifically because model access is the lever they use to justify upgrades.
Questsmith putting this on the free plan means you can run a serious campaign, with real persistent memory, D20 mechanics, live combat, and a companion system, and choose the AI that fits your story, without paying anything.
The paid tiers still exist for players who want higher memory limits, more sessions, and the full feature set. But for someone who wants to run a proper text based RPG campaign and actually test whether AI can hold a story together across sessions, the free plan is now a legitimate way to do that.
If you have been sitting on trying it because you were not sure the free tier was worth the time investment, this changes that calculation. Three AI models, no credit card, no time limit on the sessions. Open a campaign and pick the one that sounds like it fits the story you want to tell.
One More Thing Worth Noting
Sakana AI Fugu being built in Japan is not just a fun fact. It means the model was trained with different cultural influences, different storytelling traditions, and different assumptions about narrative pacing than the US-developed models.
That shows up in how it writes. If you have been running campaigns for a while and the output has started feeling predictable, like you can anticipate the structure of every scene before it generates, Fugu is genuinely disorienting in the best way. It does not write the way Western AI models write.
For players who have been using AI RPG platforms long enough to have developed a sense of what they will generate, that unpredictability is worth trying at least once.
