prompt-complexity-analyzer 0.1.0

ML-powered prompt complexity analyzer for LLM routing

complexity-analyzer

ML-powered prompt complexity analyzer for LLM routing.

Install

pip install .                        # basic (keyword features)
pip install ".[embeddings]"          # + semantic embeddings (better accuracy)
pip install ".[all]"                 # everything

Usage

from complexity_analyzer import complexity

r = complexity("prove P≠NP")
print(r.score)    # 9.2
print(r.model)    # claude-opus-4-6
print(r.tier)     # capable
print(r)          # Score 9.2/10 | Tier: capable | Model: claude-opus-4-6 | Backend: ml
r.explain()       # full breakdown with dimension scores

With semantic embeddings (requires pip install ".[embeddings]")

from complexity_analyzer import load_embedding_model, complexity

load_embedding_model()   # call once at startup — downloads ~80MB once, cached
r = complexity("your prompt")
print(r.backend)  # ml

CLI

complexity-analyzer -p "your prompt"
complexity-analyzer --only score -p "your prompt"
complexity-analyzer --provider openai -p "your prompt"
complexity-analyzer --json -p "your prompt"

Retrain with your own data

# Generate labeled data
python generate_dataset.py       # 50k examples, free APIs
python generate_targeted.py      # targeted complex examples

# Retrain
python training.py --data dataset.jsonl --balance

# Replace bundled model
cp model.joblib complexity_analyzer/model.joblib
pip install -e .                  # re-install to pick up new model

Score scale

Score	Tier	Use
1–3.5	fast	Haiku / GPT-4o-mini / Gemini Flash
3.6–6.5	balanced	Sonnet / GPT-4o / Gemini Pro
6.6–10	capable	Opus / o1 / Gemini Ultra