How we score AI devices

How capable is the underlying language model for reasoning, follow-ups, and accuracy?

0 / 10

Scripted responses, no real LLM, frequent nonsense.

5 / 10

GPT-3.5-class behaviour, often useful but hallucinates.

10 / 10

Frontier-model behaviour: nuanced reasoning, tool-use, low hallucination.

Speech recognition quality across accents, noise, and natural speech.

0 / 10

Only works in quiet rooms with American English.

5 / 10

Solid in clean audio, drops in noise or with accents.

10 / 10

Near-human ASR across accents, languages, and noisy environments.

Camera + vision model: object recognition, OCR, scene description.

0 / 10

No camera, or camera is decorative.

5 / 10

Useful for identifying common objects and reading clear text.

10 / 10

Real-time scene understanding, OCR in poor lighting, multimodal Q&A.

Does the device remember past conversations, build a useful user profile, and recall relevant context?

0 / 10

Stateless — every session starts from zero.

5 / 10

Short-term memory within a session; manual recall.

10 / 10

Long-term, queryable memory with transparent edit controls.

Can the device take real-world actions — book, send, buy, plan — on the user's behalf?

0 / 10

Read-only assistant; suggestions only.

5 / 10

A few hardcoded integrations (timer, calendar).

10 / 10

True agent behaviour with reliable multi-step task execution.

Transparency about what is recorded, where data is stored, and user control over deletion.

0 / 10

Opaque data flow, no opt-outs, foreign-jurisdiction storage.

5 / 10

Standard cloud LLM with documented retention and basic deletion.

10 / 10

On-device processing where possible, end-to-end encryption, full deletion and export controls.

Practical run time for the device's intended use pattern.

0 / 10

Cannot last a typical session of use.

5 / 10

Lasts most of a workday with moderate use.

10 / 10

Multi-day battery in real-world use, fast top-ups.

Is the device's price justified by what it actually does today?

0 / 10

Egregiously overpriced for the delivered capability.

5 / 10

Fair price for what it does, comparable to alternatives.

10 / 10

Outperforms its price tier; clear value for money.

Companion apps, third-party integrations, and pace of meaningful software updates.

0 / 10

Closed, no updates, abandoned roadmap.

5 / 10

Modest first-party app, occasional updates.

10 / 10

Active SDK, integrations, monthly meaningful updates.

Repair score, modular battery, recycled materials, and end-of-life support.

0 / 10

Glued shut, no parts, e-waste in 18 months.

5 / 10

Some user-serviceable parts, no formal sustainability program.

10 / 10

Modular, repairable, recycled materials, take-back program.