announcement and text model poll
the upcoming update, planned for around thursday, will include:
- a complete overhaul of custom contexts
- character card v2/v3 support
- a story event featuring d&d
- a new scenario set in the world of one thousand and one nights
- the ability to assign loras to individual characters
- and, of course, bug fixes and a bit of refactoring
saiga has been the default model for a little over a year now. i think it’s time to find a replacement — or make sure it still satisfies most players
the selection criteria are: it must run on laptop gpus, have between 9b and 12b parameters, support multiple languages, and be uncensored
- Qwen3.5-9B-Claude-4.6-HighIQ-THINKING-HERETIC-UNCENSORED.Q4_K_M
- Qwythos-9B-Claude-Mythos-5-1M-uncensored-heretic-Q4_K_M
- gemma-4-12B-it-uncensored-opus4.7-cot-Q4_K_M
- MN-CaptainErisNebula-Chimera-v1.1-THINKING-ClaudeOpus4.5-12B-heretic-uncensored.i1-Q4_K_M
- Mistral-Nemo-2407-12B-Thinking-Claude-Gemini-GPT5.2-Uncensored-HERETIC.i1-Q4_K_M
- saiga_nemo_12b.Q4_K_M
the first two models are built on the new qwen3.5 generation, while gemma is based on the new gemma 4. the remaining candidates are based on mistral nemo, just like the current saiga, but with more recent fine-tuning
if yr favorite model isn’t on the list, post it in the comments — including models larger than 12b. it won’t become the default model, but it may be added to the recommended list for the lucky owners of powerful hardware :):
which model is fcking better?
Qwen3.5-9B-Claude-4.6-HighIQ-THINKING-HERETIC-UNCENSORED.Q4_K_M
Qwythos-9B-Claude-Mythos-5-1M-uncensored-heretic-Q4_K_M
gemma-4-12B-it-uncensored-opus4.7-cot-Q4_K_M
1 vote
MN-CaptainErisNebula-Chimera-v1.1-THINKING-ClaudeOpus4.5-12B-heretic-uncensored.i1-Q4_K_M
Mistral-Nemo-2407-12B-Thinking-Claude-Gemini-GPT5.2-Uncensored-HERETIC.i1-Q4_K_M
saiga_nemo_12b.Q4_K_M
1 vote
2 users voted
announcement
local
local model
poll
text model
ASLPro3D ArtSimulatesLife
For the record, if you are looking for a truly uncensored Qwen, you'd do better to go with
Qwen3-8B-Hivemind-Instruct-Heretic-Abliterated-Uncensored-NEO-Imatrix-GGUF
done by DavidAU and team... lightweight (8b is only 8.14gb), VERY uncensored, virtually NO guardrails and solidly trained. I put a link on your discord.
Jul 20 18:23 (changed)
Multiic
ASLPro3D ArtSimulatesLife, thx 
Jul 20 19:38 
1
Huggtand
I love the language of Gryphe -StyleTune series. Snappy dialogue and good at multi character RP. Gryphe has redused AI clichés with 60%. Can recommend
mradermacher/Gemma-4-31B-StyleTune-heretic-ara-i1-GGUF and there is a smaller model also
mradermacher/gemma-4-12b-heretic-styletune-head-i1-GGUF
Jul 21 18:47
Huggtand
Thanks for a great game. Would it be hard to add a toggle for context SWA and ContextShift for koboldcpp in the game? So we can chose between them.
Jul 21 18:53
Mida AKhiko
Saiga seems to be working fine for me, and I don't really have any complaints other than having to occasionally make the generator remember past things. Side note: Would it be possible to add a button that is hard coded to force the end of day stuff, like the day changing over, stat calculations, and that stuff? I find in longer sessions, the game forgets how to do that, and its very immersion breaking to have to force the game to remember that hey, there's emoji triggers and I should be changing the day.
Jul 21 23:07
