server/src/db/model-pricing.ts

import type { TursoDb } from './client.js';

/**
 * Paid-equivalent pricing per model: what the SAME model (or its nearest
 * equivalent) costs per million tokens on paid APIs. Used by the analytics
 * "Est. savings" stat so it reflects realistic savings rather than pricing
 * every token like a frontier model.
 *
 * Source: OpenRouter public pricing API (paid, non-:free variants),
 * snapshot 2026-06-05; closed models use their official API prices.
 * `null` = no paid equivalent exists (stealth/preview models) — analytics
 * falls back to a modest default.
 *
 * Format: [platform, model_id, $/M input, $/M output]
 */
type PricingRow = [string, string, number | null, number | null];

export const MODEL_PRICING: PricingRow[] = [
  // Cerebras
  ['cerebras', 'gpt-oss-120b', 0.039, 0.18],
  ['cerebras', 'llama3.1-8b', 0.02, 0.03],
  ['cerebras', 'qwen-3-235b-a22b-instruct-2507', 0.071, 0.10],
  ['cerebras', 'zai-glm-4.7', 0.40, 1.75],
  // legacy ids (older DBs)
  ['cerebras', 'qwen-3-coder-480b', 0.22, 1.80],
  ['cerebras', 'llama-4-maverick-17b-128e-instruct', 0.15, 0.60],
  ['cerebras', 'qwen3-235b', 0.455, 1.82],

  // Cloudflare Workers AI
  ['cloudflare', '@cf/deepseek-ai/deepseek-r1-distill-qwen-32b', 0.29, 0.29],
  ['cloudflare', '@cf/google/gemma-4-26b-a4b-it', 0.06, 0.33],
  ['cloudflare', '@cf/ibm-granite/granite-4.0-h-micro', 0.017, 0.112],
  ['cloudflare', '@cf/meta/llama-3.3-70b-instruct-fp8-fast', 0.10, 0.32],
  ['cloudflare', '@cf/meta/llama-4-scout-17b-16e-instruct', 0.08, 0.30],
  ['cloudflare', '@cf/moonshotai/kimi-k2.6', 0.684, 3.42],
  ['cloudflare', '@cf/nvidia/nemotron-3-120b-a12b', 0.09, 0.45],
  ['cloudflare', '@cf/openai/gpt-oss-120b', 0.039, 0.18],
  ['cloudflare', '@cf/qwen/qwen3-30b-a3b-fp8', 0.09, 0.45],
  ['cloudflare', '@cf/zai-org/glm-4.7-flash', 0.06, 0.40],
  ['cloudflare', '@cf/meta/llama-3.1-70b-instruct', 0.40, 0.40], // legacy

  // Cohere (official API prices; Reasoning shares Command A pricing)
  ['cohere', 'command-a-03-2025', 2.50, 10.00],
  ['cohere', 'command-a-reasoning-08-2025', 2.50, 10.00],
  ['cohere', 'command-r-08-2024', 0.15, 0.60],
  ['cohere', 'command-r-plus-08-2024', 2.50, 10.00],

  // GitHub Models (OpenAI official prices)
  ['github', 'gpt-4o', 2.50, 10.00],
  ['github', 'openai/gpt-4.1', 2.00, 8.00],
  ['github', 'openai/gpt-5', 1.25, 10.00], // legacy

  // Google AI Studio (official prices)
  ['google', 'gemini-2.5-flash', 0.30, 2.50],
  ['google', 'gemini-2.5-flash-lite', 0.10, 0.40],
  ['google', 'gemini-2.5-pro', 1.25, 10.00],
  ['google', 'gemini-3-flash-preview', 0.50, 3.00],
  ['google', 'gemini-3.1-flash-lite-preview', 0.25, 1.50],
  ['google', 'gemini-3.1-pro-preview', 2.00, 12.00],
  ['google', 'gemini-3.5-flash', 1.50, 9.00],
  ['google', 'gemma-4-26b-a4b-it', 0.06, 0.33],
  ['google', 'gemma-4-31b-it', 0.12, 0.37],

  // Groq (compound is an agentic pipeline — estimated at its underlying
  // gpt-oss models' prices)
  ['groq', 'groq/compound', 0.039, 0.18],
  ['groq', 'groq/compound-mini', 0.029, 0.14],
  ['groq', 'llama-3.1-8b-instant', 0.02, 0.03],
  ['groq', 'llama-3.3-70b-versatile', 0.10, 0.32],
  ['groq', 'meta-llama/llama-4-scout-17b-16e-instruct', 0.08, 0.30],
  ['groq', 'llama-4-scout-17b-16e-instruct', 0.08, 0.30], // legacy id
  ['groq', 'openai/gpt-oss-120b', 0.039, 0.18],
  ['groq', 'openai/gpt-oss-20b', 0.029, 0.14],
  ['groq', 'openai/gpt-oss-safeguard-20b', 0.075, 0.30],
  ['groq', 'qwen/qwen3-32b', 0.08, 0.28],

  // Hugging Face Inference
  ['huggingface', 'Qwen/Qwen3-Coder-Next', 0.11, 0.80],
  ['huggingface', 'deepseek-ai/DeepSeek-V4-Flash', 0.098, 0.197],
  ['huggingface', 'moonshotai/Kimi-K2.6', 0.684, 3.42],
  ['huggingface', 'accounts/fireworks/models/llama-v3p3-70b-instruct', 0.10, 0.32], // legacy

  // Kilo (Poolside Laguna is stealth — no paid equivalent)
  ['kilo', 'nvidia/nemotron-3-super-120b-a12b:free', 0.09, 0.45],
  ['kilo', 'poolside/laguna-m.1:free', null, null],
  ['kilo', 'poolside/laguna-xs.2:free', null, null],
  ['kilo', 'stepfun/step-3.7-flash:free', 0.20, 1.15],

  // LLM7
  ['llm7', 'codestral-latest', 0.30, 0.90],

  // Mistral (official La Plateforme prices; Magistral per official page)
  ['mistral', 'codestral-latest', 0.30, 0.90],
  ['mistral', 'devstral-latest', 0.40, 2.00],
  ['mistral', 'magistral-medium-latest', 2.00, 5.00],
  ['mistral', 'ministral-8b-latest', 0.15, 0.15],
  ['mistral', 'mistral-large-latest', 0.50, 1.50],
  ['mistral', 'mistral-medium-latest', 1.50, 7.50],
  ['mistral', 'mistral-small-latest', 0.15, 0.60],

  // Moonshot / MiniMax (legacy platforms, may exist in older DBs)
  ['moonshot', 'kimi-latest', 0.684, 3.42],
  ['minimax', 'MiniMax-M1', 0.40, 2.20],

  // NVIDIA NIM
  ['nvidia', 'deepseek-ai/deepseek-v4-flash', 0.098, 0.197],
  ['nvidia', 'deepseek-ai/deepseek-v4-pro', 0.435, 0.87],
  ['nvidia', 'google/gemma-4-31b-it', 0.12, 0.37],
  ['nvidia', 'meta/llama-3.1-70b-instruct', 0.40, 0.40],
  ['nvidia', 'meta/llama-3.3-70b-instruct', 0.10, 0.32],
  ['nvidia', 'meta/llama-4-maverick-17b-128e-instruct', 0.15, 0.60],
  ['nvidia', 'minimaxai/minimax-m2.7', 0.279, 1.20],
  ['nvidia', 'mistralai/mistral-large-3-675b-instruct-2512', 0.50, 1.50],
  ['nvidia', 'moonshotai/kimi-k2.6', 0.684, 3.42],
  ['nvidia', 'nvidia/nemotron-3-nano-30b-a3b', 0.05, 0.20],
  ['nvidia', 'nvidia/nemotron-3-super-120b-a12b', 0.09, 0.45],
  ['nvidia', 'qwen/qwen3-coder-480b-a35b-instruct', 0.22, 1.80],
  ['nvidia', 'z-ai/glm-5.1', 0.98, 3.08],

  // Ollama (local models priced at their cloud-API equivalents — that's
  // what running them elsewhere would cost)
  ['ollama', 'cogito-2.1:671b', 1.25, 1.25],
  ['ollama', 'deepseek-v3.2', 0.229, 0.343],
  ['ollama', 'devstral-2:123b', 0.40, 2.00],
  ['ollama', 'gemma4:31b', 0.12, 0.37],
  ['ollama', 'glm-4.7', 0.40, 1.75],
  ['ollama', 'gpt-oss:120b', 0.039, 0.18],
  ['ollama', 'gpt-oss:20b', 0.029, 0.14],
  ['ollama', 'kimi-k2-thinking', 0.60, 2.50],
  ['ollama', 'mistral-large-3:675b', 0.50, 1.50],
  ['ollama', 'qwen3-coder-next', 0.11, 0.80],
  ['ollama', 'qwen3-coder:480b', 0.22, 1.80],

  // OpenCode Zen (big-pickle is stealth — no equivalent)
  ['opencode', 'big-pickle', null, null],
  ['opencode', 'deepseek-v4-flash-free', 0.098, 0.197],
  ['opencode', 'mimo-v2.5-free', 0.14, 0.28],
  ['opencode', 'nemotron-3-super-free', 0.09, 0.45],

  // OpenRouter :free pools (priced at the same model's paid variant)
  ['openrouter', 'google/gemma-4-26b-a4b-it:free', 0.06, 0.33],
  ['openrouter', 'google/gemma-4-31b-it:free', 0.12, 0.37],
  // LFM 2.5 1.2B has no paid listing; tiny-model estimate
  ['openrouter', 'liquid/lfm-2.5-1.2b-instruct:free', 0.01, 0.04],
  ['openrouter', 'liquid/lfm-2.5-1.2b-thinking:free', 0.01, 0.04],
  ['openrouter', 'meta-llama/llama-3.3-70b-instruct:free', 0.10, 0.32],
  ['openrouter', 'nousresearch/hermes-3-llama-3.1-405b:free', 1.00, 1.00],
  ['openrouter', 'nvidia/nemotron-3-nano-30b-a3b:free', 0.05, 0.20],
  ['openrouter', 'nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free', 0.05, 0.20],
  ['openrouter', 'nvidia/nemotron-3-super-120b-a12b:free', 0.09, 0.45],
  ['openrouter', 'nvidia/nemotron-nano-9b-v2:free', 0.04, 0.16],
  ['openrouter', 'openai/gpt-oss-120b:free', 0.039, 0.18],
  ['openrouter', 'openai/gpt-oss-20b:free', 0.029, 0.14],
  ['openrouter', 'openrouter/owl-alpha', null, null], // stealth
  ['openrouter', 'poolside/laguna-m.1:free', null, null],
  ['openrouter', 'poolside/laguna-xs.2:free', null, null],
  ['openrouter', 'qwen/qwen3-coder:free', 0.22, 1.80],
  ['openrouter', 'qwen/qwen3-next-80b-a3b-instruct:free', 0.09, 1.10],
  ['openrouter', 'z-ai/glm-4.5-air:free', 0.125, 0.85],
  // legacy ids
  ['openrouter', 'deepseek/deepseek-v3.1:free', 0.21, 0.79],
  ['openrouter', 'moonshotai/kimi-k2:free', 0.57, 2.30],

  // Pollinations (serves gpt-oss-20b)
  ['pollinations', 'openai-fast', 0.029, 0.14],

  // SambaNova
  ['sambanova', 'DeepSeek-V3.1', 0.21, 0.79],
  ['sambanova', 'DeepSeek-V3.2', 0.229, 0.343],
  ['sambanova', 'Llama-4-Maverick-17B-128E-Instruct', 0.15, 0.60],
  ['sambanova', 'Meta-Llama-3.3-70B-Instruct', 0.10, 0.32],
  ['sambanova', 'gemma-3-12b-it', 0.04, 0.13],
  ['sambanova', 'gpt-oss-120b', 0.039, 0.18],

  // Zhipu (4.5-flash estimated at the 4.7-flash rate — no paid 4.5-flash)
  ['zhipu', 'glm-4.5-flash', 0.06, 0.40],
  ['zhipu', 'glm-4.7-flash', 0.06, 0.40],
];

/** Fallback $/M for models with no mapping (custom endpoints, stealth). */
export const FALLBACK_INPUT_PER_M = 0.20;
export const FALLBACK_OUTPUT_PER_M = 0.80;

/**
 * Adds the pricing columns (idempotent) and refreshes prices for every
 * known model. Runs on every boot — it's ~100 UPDATEs and keeps prices
 * current when this map is updated in a release.
 */
export async function applyModelPricing(db: TursoDb): Promise<void> {
  const columns = await db.all<{ name: string }>('PRAGMA table_info(models)');
  if (!columns.some(c => c.name === 'paid_input_per_m')) {
    await db.run('ALTER TABLE models ADD COLUMN paid_input_per_m REAL');
  }
  if (!columns.some(c => c.name === 'paid_output_per_m')) {
    await db.run('ALTER TABLE models ADD COLUMN paid_output_per_m REAL');
  }

  for (const [platform, modelId, input, output] of MODEL_PRICING) {
    await db.run(
      'UPDATE models SET paid_input_per_m = ?, paid_output_per_m = ? WHERE platform = ? AND model_id = ?',
      [input, output, platform, modelId],
    );
  }
}