Model Compatibility Matrix

Which features work with which model and language?

Not every AI analysis works equally reliably with every local model. This matrix shows, based on measured benchmark data, which combinations are approved (✅), come with a latency notice (⚠️), come with an accuracy notice (◐), or are blocked (❌). Cloud models (OpenAI, Anthropic, Mistral) have no known restrictions. Apple Intelligence (On-Device) has language- and feature-specific restrictions – see matrix.

Legend

  • ✓ Approved – JSON reliability ≥ 78% and latency < 5 s (Coaching: 10 s)
  • ⚠ Latency notice – feature active, response time > 5 s (Coaching: 10 s) on the selected reference hardware
  • ✗ Blocked – JSON reliability below 78%, feature disabled
  • ◐ Accuracy notice: feature active, but rarely correct in the benchmark; the app shows a notice

Blocks and accuracy notices

Applies to all hardware configurations. Blocked when JSON reliability is below 78%; accuracy notices are individual decisions for very low hit rates. Since measuring with a JSON schema, no local model is blocked.

Qwen 2.5 1.5B

Features DEENESFRITNLPTPLSVDANOCSROJARU
Emotion Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Argumentation Fallacies ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐
Topic Classification ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Tonality Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Nonviolent Communication ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cognitive Biases ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Four-Sides Model ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Bible Impulse ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Transactional Analysis ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐ ◐
Appraisal Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Regulatory Focus ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Coaching ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Qwen 3 4B Default

Features DEENESFRITNLPTPLSVDANOCSROJARU
Emotion Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Argumentation Fallacies ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Topic Classification ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Tonality Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Nonviolent Communication ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cognitive Biases ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Four-Sides Model ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Bible Impulse ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Transactional Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Appraisal Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Regulatory Focus ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Coaching ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Qwen 2.5 7B

Features DEENESFRITNLPTPLSVDANOCSROJARU
Emotion Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Argumentation Fallacies ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Topic Classification ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Tonality Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Nonviolent Communication ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cognitive Biases ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Four-Sides Model ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Bible Impulse ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Transactional Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Appraisal Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Regulatory Focus ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Coaching ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Qwen 3 8B Premium

Features DEENESFRITNLPTPLSVDANOCSROJARU
Emotion Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Argumentation Fallacies ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Topic Classification ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Tonality Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Nonviolent Communication ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cognitive Biases ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Four-Sides Model ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Bible Impulse ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Transactional Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Appraisal Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Regulatory Focus ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Coaching ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Apple Intelligence On-Device

Features DEENESFRITNLPTPLSVDANOCSROJARU
Emotion Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✗ ✓ ✓ ✓ ✗ ✗ ✓ ✗
Argumentation Fallacies ✓ ✓ ✓ ✓ ✗ ✗ ✗ ✗ ✓ ✗ ✗ ✗ ✗ ✓ ✗
Topic Classification ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✗ ✓ ✓ ✓ ✗ ✗ ✓ ✗
Tonality Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✗ ✓ ✓ ✓ ✗ ✗ ✓ ✗
Nonviolent Communication ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✗ ✓ ✓ ✓ ✗ ✗ ✗ ✗
Cognitive Biases ✗ ✗ ✓ ✓ ✗ ✓ ✗ ✗ ✓ ✗ ✓ ✗ ✗ ✗ ✗
Four-Sides Model ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✗ ✓ ✓ ✓ ✗ ✗ ✓ ✗
Bible Impulse ✓ ✗ ✗ ✓ ✗ ✗ ✓ ✗ ✗ ✗ ✗ ✗ ✗ ✗ ✗
Transactional Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✗ ✓ ✓ ✓ ✗ ✗ ✓ ✗
Appraisal Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✗ ✓ ✓ ✓ ✗ ✗ ✓ ✗
Regulatory Focus ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✗ ✓ ✓ ✓ ✗ ✗ ✓ ✗
Coaching ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✗ ✓ ✓ ✓ ✗ ✗ ✗ ✗

◐ Correct answers in the benchmark (average across 15 languages): Qwen 2.5 1.5B, Argumentation Fallacies: 21% · Qwen 2.5 1.5B, Transactional Analysis: 46%

Latency Notices by Hardware

Latency notices vary by hardware.

– = combination blocked (see above) or no valid measurement on this hardware (older measurement without JSON schema).

Reference Hardware

🗓 2026-09-18

Qwen 2.5 1.5B

Features DEENESFRITNLPTPLSVDANOCSROJARU
Emotion Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Argumentation Fallacies ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Topic Classification ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Tonality Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Nonviolent Communication ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cognitive Biases ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Four-Sides Model ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Bible Impulse ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Transactional Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Appraisal Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Regulatory Focus ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Coaching ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Qwen 3 4B Default

Features DEENESFRITNLPTPLSVDANOCSROJARU
Emotion Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Argumentation Fallacies ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Topic Classification ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Tonality Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Nonviolent Communication ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Cognitive Biases ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Four-Sides Model ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Bible Impulse ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Transactional Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Appraisal Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Regulatory Focus ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Coaching ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓

Qwen 2.5 7B

Features DEENESFRITNLPTPLSVDANOCSROJARU
Emotion Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Argumentation Fallacies ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Topic Classification ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Tonality Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Nonviolent Communication ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ⚠ ✓ ✓ ✓
Cognitive Biases ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Four-Sides Model ✓ ✓ ✓ ⚠ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ⚠ ✓ ✓ ✓
Bible Impulse ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Transactional Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Appraisal Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Regulatory Focus ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Coaching ✓ ✓ ✓ ✓ ✓ ✓ ✓ ⚠ ⚠ ✓ ✓ ✓ ✓ ✓ ✓

Qwen 3 8B Premium

Features DEENESFRITNLPTPLSVDANOCSROJARU
Emotion Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Argumentation Fallacies ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Topic Classification ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Tonality Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Nonviolent Communication ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ⚠ ⚠ ✓ ✓
Cognitive Biases ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Four-Sides Model ✓ ✓ ✓ ✓ ⚠ ⚠ ✓ ⚠ ✓ ✓ ✓ ⚠ ⚠ ✓ ✓
Bible Impulse ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Transactional Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Appraisal Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Regulatory Focus ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Coaching ✓ ✓ ✓ ✓ ✓ ✓ ✓ ⚠ ⚠ ✓ ✓ ⚠ ⚠ ✓ ✓

Apple Intelligence On-Device

Features DEENESFRITNLPTPLSVDANOCSROJARU
Emotion Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ – ✓ ✓ ✓ – – ✓ –
Argumentation Fallacies ✓ ✓ ✓ ✓ – – – – ✓ – – – – ✓ –
Topic Classification ✓ ✓ ✓ ✓ ✓ ✓ ✓ – ✓ ✓ ✓ – – ✓ –
Tonality Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ – ✓ ✓ ✓ – – ✓ –
Nonviolent Communication ✓ ✓ ✓ ✓ ✓ ✓ ✓ – ✓ ✓ ✓ – – – –
Cognitive Biases – – ✓ ✓ – ✓ – – ✓ – ✓ – – – –
Four-Sides Model ✓ ✓ ✓ ✓ ✓ ✓ ✓ – ✓ ✓ ✓ – – ✓ –
Bible Impulse ✓ – – ✓ – – ✓ – – – – – – – –
Transactional Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ – ✓ ✓ ✓ – – ✓ –
Appraisal Analysis ✓ ✓ ✓ ✓ ✓ ✓ ✓ – ✓ ✓ ✓ – – ✓ –
Regulatory Focus ✓ ✓ ✓ ✓ ✓ ✓ ✓ – ✓ ✓ ✓ – – ✓ –
Coaching ⚠ ✓ ✓ ⚠ ⚠ ⚠ ⚠ – ⚠ ⚠ ⚠ – – – –

☁️ Cloud Models (OpenAI, Anthropic, Mistral)

All 12 analyses in all 15 languages approved – no known restrictions.

Measurement Criteria

Go/No-Go per ADR-075: JSON reliability ≥ 78% and median latency (p50) < 5,000 ms (Coaching: 10 s). Reference hardware: Apple M4 Pro (14C CPU + 20C GPU, 64 GB RAM), NVIDIA RTX 2000 Ada (16 GB GDDR6, CUDA 13.1), Apple M1 and Ryzen 5 (CPU). 10 runs per combination; emotion, fallacy detection and transactional analysis with 10 test texts per language and 20 runs.