Model reference · open weights

Mistral-Trismegistus

LLMs teknium Text gen 1 build Open weights 67 dl/mo

Mistral-Trismegistus is an open-weight language model from teknium. Mistral-Trismegistus-7B (BF16) weighs 14.5 GB; the smallest configuration that runs it is 2× RTX 3060 12 GB.

  • Mistral-Trismegistus is a 7B parameter text-generation model created by teknium and published as teknium/Mistral-Trismegistus-7B.
  • It is designed for tasks related to the esoteric, occult, and spiritual, utilizing a synthetic dataset of approximately 10,000 examples.
  • The model supports a context length of 32,768 tokens, operates in English, and is distributed under the Apache 2.0 license.

Summary of the teknium/Mistral-Trismegistus-7B model card, 2026-10-01

What it is

Released byteknium
Released2023-10-07
VRAM14.5 GB for the weights

What it runs on

Memory and cards for Mistral-Trismegistus-7B (BF16)

14.5 GBweights, file size
0 MBcache per 1K tokens
1.1 GBwindow cache per request
647 MBruntime overhead, at least
32,768 tokenscontext max
CardRequests at onceContext maxMemory
8K each32K each
RTX 3060 12 GB———11.6 GB
RTX 4060 Ti 16 GB———15.4 GB
RTX 3090 24 GB77all 32K23.4 GB
RTX 4090 24 GB77all 32K23.4 GB
RTX 5090 32 GB1414all 32K31.0 GB
L40S 48 GB2626all 32K44.0 GB
A100 80 GB5858all 32K78.2 GB
H100 80 GB5522all 32K78.1 GB
RTX PRO 6000 Blackwell 96 GB7028all 32K93.8 GB
DGX Spark (GB10) 128 GB unified8333all 32K107 GB
H200 141 GB11144all 32K138 GB
B200 180 GB14636all 32K176 GB
2× RTX 3060 12 GB
tensor parallel
66all 32K11.6 GB a card
2× RTX 4060 Ti 16 GB
tensor parallel
1414all 32K15.4 GB a card
2× RTX 4090 24 GB
tensor parallel
2828all 32K23.4 GB a card
2× RTX 3090 24 GB
tensor parallel
2828all 32K23.4 GB a card
Memory needed at each load
Requests at once8K tokens each32K tokens each
116.2 GB16.2 GB
520.5 GB20.5 GB
823.7 GB23.7 GB
1632.3 GB32.3 GB
3249.5 GB49.5 GB
6483.8 GB83.8 GB

One card, with vLLM's small-card settings.

From the model card

What teknium says about Mistral-Trismegistus

Read the model card

Mistral Trismegistus 7B

Model Description:

Transcendence is All You Need! Mistral Trismegistus is a model made for people interested in the esoteric, occult, and spiritual.

Here are some outputs:

Answer questions about occult artifacts:

Play the role of a hypnotist:

Special Features:

  • The First Powerful Occult Expert Model: ~10,000 high quality, deep, rich, instructions on the occult, esoteric, and spiritual.
  • Fast: Trained on Mistral, a state of the art 7B parameter model, you can run this model FAST on even a cpu.
  • Not a positivity-nazi: This model was trained on all forms of esoteric tasks and knowledge, and is not burdened by the flowery nature of many other models, who chose positivity over creativity.

Acknowledgements:

Special thanks to @a16z.

Dataset:

This model was trained on a 100% synthetic, gpt-4 generated dataset, about ~10,000 examples, on a wide and diverse set of both tasks and knowledge about the esoteric, occult, and spiritual.

The dataset will be released soon!

Usage:

Prompt Format:

USER:
ASSISTANT:

OR

USER:
ASSISTANT:

Benchmarks:

No benchmark can capture the nature and essense of the quality of spirituality and esoteric knowledge and tasks. You will have to try testing it yourself!

Training run on wandb here: https://wandb.ai/teknium1/occult-expert-mistral-7b/runs/coccult-expert-mistral-6/overview

Licensing:

Apache 2.0


Quoted from the model card on Hugging Face. The full card is behind the Hugging Face link above.

© 2026 AxForge · EU-hosted AI infrastructure Pricing Docs Trust Privacy Terms