version: "cgn.v1" schema_version: "1" effective_date: "2026-05-14" status: proposed description: "Canonical CGN conversion profiles for model-token accounting across providers." formula: id: cgn.v1 expression: "ceil(((input_tokens * input_weight) + (output_tokens * output_weight) + (thinking_tokens * thinking_weight)) * model_coefficient / scale)" missing_token_class_value: 0 rounding: mode: ceil output_type: uint32 defaults: input_weight: 1 output_weight: 4 thinking_weight: 2 scale: 1000 model_coefficient: 1 unknown_model_policy: profile_id: default.unknown warn: true telemetry_flag: cgn_profile_defaulted billing_allowed: false behavior: "Unknown models use default.unknown for CGN-Estimate. CGN-Billing has no unknown-model fallback." profiles: - id: default.unknown provider: default model_patterns: ["*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 1 scale: 1000 billing_allowed: false - id: ds.deepseek-chat provider: ds model_patterns: ["deepseek-chat", "deepseek-chat-*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 1 scale: 1000 billing_allowed: true - id: ds.deepseek-reasoner provider: ds model_patterns: ["deepseek-reasoner", "deepseek-reasoner-*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 2 scale: 1000 billing_allowed: true - id: oa.gpt-4o provider: oa model_patterns: ["gpt-4o", "gpt-4o-*", "gpt-4.1", "gpt-4.1-*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 1.5 scale: 1000 billing_allowed: true - id: oa.reasoning provider: oa model_patterns: ["o1", "o1-*", "o3", "o3-*", "o4-mini", "o4-mini-*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 3 scale: 1000 billing_allowed: true - id: an.claude-haiku provider: an model_patterns: ["claude-*-haiku-*", "*haiku*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 0.75 scale: 1000 billing_allowed: true - id: an.claude-sonnet provider: an model_patterns: ["claude-*-sonnet-*", "*sonnet*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 2 scale: 1000 billing_allowed: true - id: an.claude-opus provider: an model_patterns: ["claude-*-opus-*", "*opus*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 4 scale: 1000 billing_allowed: true - id: go.gemini-flash provider: go model_patterns: ["gemini-*-flash*", "gemini-2.0-flash*", "gemini-2.5-flash*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 0.75 scale: 1000 billing_allowed: true - id: go.gemini-pro provider: go model_patterns: ["gemini-*-pro*", "gemini-1.5-pro*", "gemini-2.5-pro*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 2 scale: 1000 billing_allowed: true - id: go.gemini-ultra provider: go model_patterns: ["gemini-ultra*", "gemini-2.0-ultra*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 4 scale: 1000 billing_allowed: true - id: meta.llama-small provider: meta model_patterns: ["llama-3*-8b*", "llama-3*-8b*", "llama3*8b*", "llama-3.2*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 0.5 scale: 1000 billing_allowed: true - id: meta.llama-mid provider: meta model_patterns: ["llama-3*-70b*", "llama3*70b*", "llama-3.1-70b*", "llama-3.3*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 1 scale: 1000 billing_allowed: true - id: meta.llama-large provider: meta model_patterns: ["llama-3*-405b*", "llama3*405b*", "llama-3.1-405b*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 2 scale: 1000 billing_allowed: true - id: mistral.small provider: mistral model_patterns: ["mistral-small*", "mistral-7b*", "open-mistral-7b*", "ministral*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 0.5 scale: 1000 billing_allowed: true - id: mistral.medium provider: mistral model_patterns: ["mistral-medium*", "mistral-8x7b*", "mixtral*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 1 scale: 1000 billing_allowed: true - id: mistral.large provider: mistral model_patterns: ["mistral-large*", "mistral-2*", "codestral*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 1.5 scale: 1000 billing_allowed: true - id: ollama.default provider: ollama model_patterns: ["*"] input_weight: 1 output_weight: 4 thinking_weight: 2 model_coefficient: 1 scale: 1000 billing_allowed: false test_vectors: - { profile_id: default.unknown, input_tokens: 1000, output_tokens: 100, thinking_tokens: 0, weighted_tokens: 1400, raw_cgn: 1.4, expected_cgn: 2 } - { profile_id: ds.deepseek-chat, input_tokens: 2000, output_tokens: 500, thinking_tokens: 0, weighted_tokens: 4000, raw_cgn: 4, expected_cgn: 4 } - { profile_id: ds.deepseek-reasoner, input_tokens: 2000, output_tokens: 500, thinking_tokens: 1200, weighted_tokens: 6400, raw_cgn: 12.8, expected_cgn: 13 } - { profile_id: oa.gpt-4o, input_tokens: 1200, output_tokens: 300, thinking_tokens: 0, weighted_tokens: 2400, raw_cgn: 3.6, expected_cgn: 4 } - { profile_id: oa.reasoning, input_tokens: 1200, output_tokens: 300, thinking_tokens: 800, weighted_tokens: 4000, raw_cgn: 12, expected_cgn: 12 } - { profile_id: an.claude-haiku, input_tokens: 1000, output_tokens: 250, thinking_tokens: 0, weighted_tokens: 2000, raw_cgn: 1.5, expected_cgn: 2 } - { profile_id: an.claude-sonnet, input_tokens: 1000, output_tokens: 250, thinking_tokens: 500, weighted_tokens: 3000, raw_cgn: 6, expected_cgn: 6 } - { profile_id: an.claude-opus, input_tokens: 1000, output_tokens: 250, thinking_tokens: 500, weighted_tokens: 3000, raw_cgn: 12, expected_cgn: 12 }