| Model | Engine | LLM calls | External APIs | Cost/msg |
|---|
| charles-auma (arithmetic) | regex → npdollars circuit | 0 | npdollars | $0 |
| charles-auma (general) | 1-2 Haiku + AUMA + npdollars + snakebatch | 1-2 | auma + npdollars + snakebatch | ~$0.003 |
| moncey (high conf) | snake/comprendre → Haiku | 1 | snake.aws | ~$0.002 |
| moncey (low conf) | snake + moncesuite (10 classifiers) → Haiku | 1 | snake.aws + moncesuite.aws | ~$0.003 |
| charles | 3x Haiku + Sonnet | 4 | charles.aws, sudoku.aws | ~$0.01 |
| concise | charles + Haiku | 5 | same as charles | ~$0.01 |
| charles-science | 2x Haiku + Snake + service + Sonnet | 3 | auma/npdollars/sudoku/chess/snake | ~$0.01 |
| charles-json | Sonnet | 1 | charles.aws (5s cap) | ~$0.01 |
| charles-architect | Sonnet | 1 | charles.aws | ~$0.01 |
| cc | charles + concise + Haiku | 10 | charles.aws x2 | ~$0.02 |
| Haiku 4.5 | Bedrock Converse | 1 | — | ~$0.003 |
| Sonnet 4.6 | Bedrock Converse | 1 | — | ~$0.03 |
| Nova Pro | Bedrock Converse | 1 | — | ~$0.008 |
| Nova Lite | Bedrock Converse | 1 | — | ~$0.001 |
| Nova Micro | Bedrock Converse | 1 | — | ~$0.0005 |
| Component | Cost | What |
|---|
| Binary multiplier circuit | $0 | regex → CNF → Kissat. 65535×65535 in 58ms, 0 tokens |
| Binary divider circuit | $0 | Q×B+R=A as SAT. 1000000÷333 in 48ms, 0 tokens |
| NP Calculator | $0 | Python Decimal, AST-safe eval |
| Snake SAT Router | $0 | Local model, 0ms, routes charles-science |
| AUMA API | $0 | CPU-only Fourier optimization |
| SnakeBatch /stream/train | $0 | Train classifier from Haiku CSV, no auth |
At 100 users × 10 msgs/day = 1000 msgs/day. Bedrock cost ~$3-10/day depending on model mix.
Total: ~$0.60/month infra + ~$100-300/month Bedrock for 1000 daily active conversations.