Picking the Right AI Model for Biggly Bot
If you haven’t already, skim our Beginner’s Guide to Local AI first. It’ll save you a lot of head-scratching.
Picking the “best” model is like picking the best book. Some people love verbose, flowing prose. Others will call that “GPT slop” and demand something tighter. There’s no single right answer, and half the fun is experimenting. That said, here’s what we’ve found works brilliantly.
Where to Find Models
The place to browse the latest roleplay-friendly, reasonably uncensored models is the UGI Leaderboard. It’s community-maintained and focuses on the kind of creative, unrestricted models that work best for character chat.
Regardless of which model catches your eye, look for these words in the name for maximum fun:
- abliterated (safety filters removed)
- heretic (ditto, different method)
- derestricted (you get the idea)
There are quite a few models marked “Uncensored” which are in fact, censored. Models without these labels above will often refuse to engage with romance, conflict, or anything remotely edgy. You didn’t buy a private, local AI app like Biggly to be told “I can’t do that.”
Our Current Favourite: The Tiger Gemma Range
I’ve personally come to love the Tiger range of Gemma models. They’re up for anything, quite chatty, happy to carry a story forward, and surprisingly intelligent for their size.
The difference between any 7B or 8B model and Tiger-Gemma-9B-v3 is dramatic. If your machine can run a 9B Gemma at an acceptable speed, I’d say go for it. Download the Q5_K_M quant for the best balance of quality and VRAM usage, or Q4_K_S if the Q5_K_M is too slow for you.
(Remember to set the chat template in Biggly Bot’s AI settings to “Gemma/Mistral” when using Gemma models. The app will suggest this automatically if it spots “gemma” in the filename, but worth knowing.)
Scaling Up by GPU
Got a mid-range card with 8-10GB VRAM? The Gemma 3 9B is the sweet spot. Genuinely impressive for its size.
Got 12GB or more? You could try the Gemma 3 12B, though honestly the 9B is so good I didn’t notice a huge leap.
Got a 90-series card (3090, 4090, 5090)?
Then Gemma 3 27B is fast, smart, and the Tiger or abliterated versions are up for absolutely anything! A firm favourite: Big-Tiger-Gemma-27B-v1-GGUF
A newer, and very smart but smaller model ideal for the Sophe character is this mouthful:
Qwen3-14B-Claude-4.5-Opus-High-Reasoning-Distill-GGUF/tree/main
A Note on the Memory Miner
Biggly Bot has a built-in memory system that reads your conversations and extracts facts your characters can remember across sessions. This “miner” benefits from a different kind of model than the one you chat with. Plain ‘Instruct’-tuned models are often a better choice here. You can set a completely separate model for mining in the character settings, so your creative chat model and your analytical miner model can each do what they’re best at. A generally good all-rounder, by our good friend Bartowski is:
Recommended Online Models (via OpenRouter)
If you’re using cloud models through Biggly’s Cloud option, such as OpenRouter or similar instead of (or alongside) running locally, these are worth trying. Online models are much larger and smarter than anything you can run at home, and you only pay for what you use, rather than a subscription.
- Hermes 4 405B – Big, capable, and willing. A reliable all-rounder.
- Mistral Large (mistralai/mistral-large-2512) – Fast, smart, and less likely to refuse creative scenarios than most cloud models.
- Kimi 2.5 Thinking – Impressive depth. Turn thinking tabs off in Biggly Bot’s UI settings or you’ll see its internal reasoning in every reply. Some say the thinking model is less censored.
- DeepSeek V3 – Arguably the best value going. Affordable, fast, deep, and remarkably unrestricted for a cloud model. The world of AI models moves fast, but this one has staying power.
The “Best Local Model for Normal PCs” Pick
If you’ve got a 90-series card and want to push the boundary of what runs locally, this 70B model is worth the slower speed. You’ll need a smaller quant (e.g. Q3_K_S) and limited context to fit it, but the quality leap over small models is real:
Getting Started is Easy
Biggly Bot handles all the fiddly bits. Download a GGUF file, point the app at it, and you’re chatting in under a minute. The first-run wizard walks you through everything, and the app auto-detects your model type to suggest the right template. No command lines, no Python installs, no “compile from source.” Just characters, stories, and your imagination!
Don’t have Biggly Bot yet? You can download a 7-Day Free Trial