Hopp til hovedinnhold
tenki
All models
The quick oneExperimental

Munin

Answers in the moment, without the long deliberation. Named for the raven that stands for memory.

· Specifications

Model
Our own Norwegian fine-tune, trained by Tenki
Size
4 billion parameters, 4-bit quantised MLX build
Context window
262,144 tokens native, practical limit lower on 16 GB
Runs on
Mac mini M4 running mlx_lm.server, located in Norway
Interface
OpenAI-compatible chat completions
Access
Free and anonymous at /chat, or with a token via the API
Conversation storage
None. Conversations are not stored.
Profile
Short reasoning, fast answers
01

What Munin is

Munin is the same fine-tuned model as Hugin, run for speed. It spends less time thinking before it starts answering, and suits you when you want an answer immediately rather than a considered one.

It shares weights and hardware with Hugin. The difference is in the configuration.

02

Intended use

Munin is for the short exchanges: quick questions, rewordings, clarifications. When the question needs the model to actually think it through, switch to Hugin.

  • · Short questions where waiting is worse than precision
  • · Rewordings and summaries you read over yourself
  • · Testing how a local model feels in practice
03

Out of scope

The same applies as for Hugin, with a slightly wider margin: Munin spends less time thinking, so the mistakes arrive faster.

  • · Legal, medical, or financial advice
  • · Processing personal data or client data
  • · Decisions about hiring, credit, health, or benefits
  • · Anything that will stand without a human reading it over
04

Limitations

Munin is a small model. Four billion parameters, quantised to four bits to fit the hardware we run it on. That is a deliberate trade: we prioritise the model sitting in Norway on hardware we own over it being as large as possible. The fine-tune recovers some ground on Norwegian, but it does not undo the size. The consequence is that it loses to the large cloud models on tasks needing broad factual knowledge, long chains of reasoning, or precise code.

  • · It makes things up. Verify anything that will be used for something that matters.
  • · Its knowledge is general and has a cutoff. It does not know your internal documents.
  • · Norwegian bokmål is weaker than English, and nynorsk weaker still.
  • · Arithmetic, dates, and statutory references are among the weakest areas.
  • · It remembers nothing between conversations, because we do not store them.
05

Data and privacy

Identical to Hugin: conversations are not stored, there is no account and no history. Aggregate usage and any thumbs up or down are stored. Rate limiting per IP address, with Turnstile in front.

06

Evaluation

The same benchmark as Hugin. The tasks, weighting, and methods are public on TenkiBench, and we measure ourselves with the same tasks we measure everyone else with.