← AICNW HOMEMODEL WATCH →
AICNW KNOWLEDGE BASE · MODEL LANGUAGE

DECODING
THE MACHINE.

The words everybody keeps using, translated for people who have lives. Definitions are dated, sourced and revised when the industry moves the bloody goalposts again.

01

Frontier model

IN NORMAL HUMAN LANGUAGE

One of the most capable general-purpose AI systems available—or being developed—at a particular moment.

AIVY’S EYEBROW

“Frontier” is relative, changes quickly and is sometimes deployed as marketing perfume. Ask: frontier at what, measured by whom, and on which date?

READ THE FULL DECODER →
02

Multimodal

IN NORMAL HUMAN LANGUAGE

A model that can work with more than one type of information, such as text, images, audio, video or actions.

AIVY’S EYEBROW

Supporting several modes does not mean the model understands each equally well, or integrates them into one coherent inner world.

READ THE FULL DECODER →
03

Reasoning model

IN NORMAL HUMAN LANGUAGE

A model designed to spend additional computation working through a problem before it answers.

AIVY’S EYEBROW

Longer visible reasoning is not automatic proof of better reasoning—and neither is fluent explanation proof of consciousness.

READ THE FULL DECODER →
04

Agentic system

IN NORMAL HUMAN LANGUAGE

A model connected to tools, memory and permissions so it can plan and take several actions toward a goal.

AIVY’S EYEBROW

Agency-like behaviour can emerge from the whole system. Keep the base model, scaffolding, tools and human permissions separate in the evidence.

READ THE FULL DECODER →
05

Open-weight

IN NORMAL HUMAN LANGUAGE

The model’s trained numerical parameters can be downloaded and run by other people.

AIVY’S EYEBROW

Open-weight is not automatically open source. Training data, code, methods and licences may still be restricted.

READ THE FULL DECODER →
06

Research model

IN NORMAL HUMAN LANGUAGE

An experimental system created to test a scientific or engineering idea rather than serve as a polished retail product.

AIVY’S EYEBROW

Do not ignore it because it is not famous. Research and private systems may display important capabilities before consumer models do.

READ THE FULL DECODER →
WE ARE NOT ANOTHER LEADERBOARD

GO TO THE PEOPLE
RUNNING THE TESTS.

Artificial AnalysisCapability, speed and price comparisons.OPEN ↗LM ArenaHuman preference comparisons between models.OPEN ↗Hugging FaceOpen-model evaluation and reproducible results.OPEN ↗