feat: add MiniMax Flash and M3 pricing fallbacks

lib/pricing.js

@@ -0,0 +1,65 @@
+// ── LLM Pricing (via @pydantic/genai-prices) ──
+const { calcPrice } = require('@pydantic/genai-prices');
+
+// Free models (no cost regardless of pricing data)
+const FREE_MODELS = new Set([
+  'xiaomi/mimo-v2-pro',
+  'mimo-v2-pro',
+]);
+
+// Hermes billing_provider → genai-prices providerId mapping
+const PROVIDER_MAP = {
+  openrouter: 'openrouter',
+  'openai-codex': 'openai',
+  'opencode-go': 'openrouter', // minimax available via OpenRouter pricing
+};
+
+// Custom pricing for models not in genai-prices (per million tokens).
+// MiniMax-M3 uses the standard <=512k baseline; the published long-context
+// threshold has no exact integer expansion, so the fallback avoids guessing it.
+const CUSTOM_PRICING = {
+  'minimax-m2': { input_mtok: 0.30, output_mtok: 1.20, cache_read_mtok: 0.03 },
+  'minimax-m2.7': { input_mtok: 0.30, output_mtok: 1.20, cache_read_mtok: 0.06 },
+  'minimax-m3.1-flash': { input_mtok: 0.15, output_mtok: 0.60, cache_read_mtok: 0.015 },
+  'minimax-m3': { input_mtok: 0.30, output_mtok: 1.20, cache_read_mtok: 0.06 },
+};
+
+// Calculate cost in USD from token counts
+function calculateCost(model, inputTokens, outputTokens, cacheReadTokens = 0, billingProvider) {
+  if (!model || FREE_MODELS.has(model)) return 0;
+
+  // genai-prices expects input_tokens = total input (including cache)
+  // Our DB stores input and cache separately, so combine them
+  const usage = {
+    input_tokens: inputTokens + cacheReadTokens,
+    output_tokens: outputTokens,
+    cache_read_tokens: cacheReadTokens,
+  };
+
+  // 1. Try with provider hint
+  const providerId = PROVIDER_MAP[billingProvider];
+  if (providerId) {
+    try {
+      const r = calcPrice(usage, model, { providerId });
+      if (r) return r.total_price;
+    } catch (_) { /* fallback to next */ }
+  }
+
+  // 2. Try without provider (match across all providers)
+  try {
+    const r = calcPrice(usage, model);
+    if (r) return r.total_price;
+  } catch (_) { /* fallback to custom pricing */ }
+
+  // 3. Custom pricing fallback
+  const custom = CUSTOM_PRICING[String(model).toLowerCase()];
+  if (custom) {
+    return (inputTokens / 1e6) * custom.input_mtok
+      + (outputTokens / 1e6) * custom.output_mtok
+      + (cacheReadTokens / 1e6) * (custom.cache_read_mtok || custom.input_mtok * 0.1);
+  }
+
+  return 0;
+}
+
+module.exports = { calculateCost };

server.js

@@ -21,65 +21,7 @@ const {
 // ── TUI Gateway Bridge ──
 const { getBridge, killAllBridges } = require('./lib/tui-gateway-bridge');
 
-// ── LLM Pricing (via @pydantic/genai-prices) ──
-const { calcPrice } = require('@pydantic/genai-prices');
-
-// Free models (no cost regardless of pricing data)
-const FREE_MODELS = new Set([
-  'xiaomi/mimo-v2-pro',
-  'mimo-v2-pro',
-]);
-
-// Hermes billing_provider → genai-prices providerId mapping
-const PROVIDER_MAP = {
-  'openrouter':    'openrouter',
-  'openai-codex':  'openai',
-  'opencode-go':   'openrouter', // minimax available via OpenRouter pricing
-};
-
-// Custom pricing for models not in genai-prices (per million tokens)
-const CUSTOM_PRICING = {
-  'minimax-m2':   { input_mtok: 0.30, output_mtok: 1.20, cache_read_mtok: 0.03 },
-  'minimax-m2.7': { input_mtok: 0.30, output_mtok: 1.20, cache_read_mtok: 0.03 },
-};
-
-// Calculate cost in USD from token counts
-function calculateCost(model, inputTokens, outputTokens, cacheReadTokens = 0, billingProvider) {
-  if (!model || FREE_MODELS.has(model)) return 0;
-
-  // genai-prices expects input_tokens = total input (including cache)
-  // Our DB stores input and cache separately, so combine them
-  const usage = {
-    input_tokens: inputTokens + cacheReadTokens,
-    output_tokens: outputTokens,
-    cache_read_tokens: cacheReadTokens,
-  };
-
-  // 1. Try with provider hint
-  const providerId = PROVIDER_MAP[billingProvider];
-  if (providerId) {
-    try {
-      const r = calcPrice(usage, model, { providerId });
-      if (r) return r.total_price;
-    } catch (_) { /* fallback to next */ }
-  }
-
-  // 2. Try without provider (match across all providers)
-  try {
-    const r = calcPrice(usage, model);
-    if (r) return r.total_price;
-  } catch (_) { /* fallback to custom pricing */ }
-
-  // 3. Custom pricing fallback
-  const custom = CUSTOM_PRICING[model];
-  if (custom) {
-    return (inputTokens / 1e6) * custom.input_mtok
-      + (outputTokens / 1e6) * custom.output_mtok
-      + (cacheReadTokens / 1e6) * (custom.cache_read_mtok || custom.input_mtok * 0.1);
-  }
-
-  return 0;
-}
+const { calculateCost } = require('./lib/pricing');
 
 // Async shell execution utility (non-blocking)
 function parseShellTimeout(value) {

test/pricing.test.js

@@ -0,0 +1,58 @@
+const test = require('node:test');
+const assert = require('node:assert/strict');
+const Module = require('node:module');
+
+function loadPricing(calcPrice) {
+  const originalLoad = Module._load;
+  Module._load = function load(request, parent, isMain) {
+    if (request === '@pydantic/genai-prices') return { calcPrice };
+    return originalLoad.call(this, request, parent, isMain);
+  };
+
+  try {
+    delete require.cache[require.resolve('../lib/pricing')];
+    return require('../lib/pricing');
+  } finally {
+    Module._load = originalLoad;
+  }
+}
+
+test('MiniMax M3.1 Flash fallback uses the documented direct rates and provider hint', () => {
+  const calls = [];
+  const { calculateCost } = loadPricing((usage, model, options) => {
+    calls.push({ usage, model, options });
+    return null;
+  });
+
+  assert.equal(
+    calculateCost('MiniMax-M3.1-Flash', 1_000_000, 1_000_000, 1_000_000, 'opencode-go'),
+    0.765
+  );
+  assert.deepEqual(calls, [
+    {
+      usage: { input_tokens: 2_000_000, output_tokens: 1_000_000, cache_read_tokens: 1_000_000 },
+      model: 'MiniMax-M3.1-Flash',
+      options: { providerId: 'openrouter' },
+    },
+    {
+      usage: { input_tokens: 2_000_000, output_tokens: 1_000_000, cache_read_tokens: 1_000_000 },
+      model: 'MiniMax-M3.1-Flash',
+      options: undefined,
+    },
+  ]);
+});
+
+test('MiniMax M3 fallback stays on the documented standard baseline', () => {
+  const { calculateCost } = loadPricing(() => null);
+
+  assert.equal(calculateCost('minimax-m3', 1_000_000, 1_000_000, 1_000_000), 1.56);
+});
+
+test('SDK pricing remains authoritative and legacy/free/unknown behavior is preserved', () => {
+  const { calculateCost: calculateWithSdk } = loadPricing(() => ({ total_price: 7.25 }));
+  assert.equal(calculateWithSdk('minimax-m3.1-flash', 1_000_000, 1_000_000), 7.25);
+
+  const { calculateCost: calculateWithFallback } = loadPricing(() => null);
+  assert.equal(calculateWithFallback('mimo-v2-pro', 1_000_000, 1_000_000), 0);
+  assert.equal(calculateWithFallback('custom/saved-model', 1_000_000, 1_000_000), 0);
+});