slash-tokens 1.5.6 → 1.5.7

package/dist/cli.js

@@ -208,8 +208,8 @@ var MODELS = {
   "claude-opus-4.7": { input: 5, output: 25, context: 1e6 },
   "claude-sonnet": { input: 2, output: 10, context: 1e6 },
   "claude-haiku": { input: 1, output: 5, context: 200000 },
-  "grok-4.20": { input: 1.25, output: 2.5, context: 1e6 },
-  "grok-4-1-fast": { input: 1.25, output: 2.5, context: 1e6 },
+  "grok-4.20": { input: 1.25, output: 2.5, context: 1e6, longContextThreshold: 200000, longContextInput: 2.5, longContextOutput: 5 },
+  "grok-4-1-fast": { input: 1.25, output: 2.5, context: 1e6, longContextThreshold: 200000, longContextInput: 2.5, longContextOutput: 5 },
   "gemini-3.1-pro": { input: 2, output: 12, context: 1e6 },
   "gemini-2.5-flash": { input: 0.3, output: 2.5, context: 1e6 },
   "gpt-5.4": { input: 2.5, output: 15, context: 1e6 },

package/dist/intercept.d.ts

@@ -11,5 +11,6 @@ export interface InterceptEvent {
     routed: boolean;
     timestamp: string;
 }
+export declare function normalizeModel(raw: string): string;
 export declare function onIntercept(handler: (event: InterceptEvent) => void): void;
 export declare function patchFetch(): void;

package/dist/intercept.js

@@ -6,6 +6,7 @@ import { PROVIDER_MODELS } from './providers.js';
 // (what to put back in the request body)
 const MODEL_API_NAMES = {
     'claude-opus': 'claude-opus-5',
+    'claude-opus-4.7': 'claude-opus-4-7',
     'claude-sonnet': 'claude-sonnet-5',
     'claude-haiku': 'claude-haiku-4-5-20251001',
     'gpt-5.4': 'gpt-5.4',
@@ -43,10 +44,20 @@ const AI_ENDPOINTS = [
         modelExtractor: (body) => body?.model || 'grok-4.20',
     },
 ];
-// Normalize model names to our pricing table keys
-function normalizeModel(raw) {
+// Normalize model names to our pricing table keys.
+// Kept in sync by hand with mcpaas-cf/src/slash-models.ts's normalizeModel
+// (no shared import between the two repos) — fixed 2026-08-23: this copy
+// was missing the opus-4.7 special case and several OpenAI legacy
+// fallbacks the other copy already had, which silently mapped those
+// intercepted calls to generic claude-opus / an unmapped raw string
+// (getModel() returning undefined → $0 reported cost, see slash.ts's
+// DEFAULT_UNKNOWN_MODEL_FACTOR comment for why "unrecognized" defaulting
+// to a falsely-safe-looking value is the dangerous case).
+export function normalizeModel(raw) {
     const lower = raw.toLowerCase();
-    // Anthropic
+    // Anthropic — 4.7 before generic opus check, same order as slash-models.ts
+    if (lower.includes('opus') && (lower.includes('4-7') || lower.includes('4.7')))
+        return 'claude-opus-4.7';
     if (lower.includes('opus'))
         return 'claude-opus';
     if (lower.includes('sonnet'))
@@ -63,20 +74,28 @@ function normalizeModel(raw) {
         return 'gemini-3.1-pro';
     if (lower.includes('gemini'))
         return 'gemini-2.5-flash';
-    // OpenAI
+    // OpenAI — current (5.4 family)
     if (lower.includes('5.4') && lower.includes('nano'))
         return 'gpt-5.4-nano';
     if (lower.includes('5.4') && lower.includes('mini'))
         return 'gpt-5.4-mini';
     if (lower.includes('5.4'))
         return 'gpt-5.4';
-    // Legacy fallbacks
+    // OpenAI — legacy model names → map to closest current equivalent
+    if (lower.includes('o1-mini') || lower.includes('o1_mini'))
+        return 'gpt-5.4-mini';
+    if (lower.includes('o1'))
+        return 'gpt-5.4';
     if (lower.includes('4o-mini') || lower.includes('4o_mini'))
         return 'gpt-5.4-mini';
     if (lower.includes('4o') || lower.includes('4.1'))
         return 'gpt-5.4';
-    if (lower.includes('grok-3'))
-        return 'grok-4.20';
+    if (lower.includes('gpt-4-turbo') || lower.includes('gpt-4-1'))
+        return 'gpt-5.4';
+    if (lower.includes('gpt-4'))
+        return 'gpt-5.4';
+    if (lower.includes('gpt-3.5') || lower.includes('gpt-35'))
+        return 'gpt-5.4-nano';
     return raw;
 }
 // Extract message content from request body

package/dist/models.d.ts

@@ -2,7 +2,14 @@ export interface ModelInfo {
     input: number;
     output: number;
     context: number;
+    longContextThreshold?: number;
+    longContextInput?: number;
+    longContextOutput?: number;
 }
 export declare const MODELS: Record<string, ModelInfo>;
 export declare function getModel(name: string): ModelInfo | undefined;
+export declare function effectiveRate(tokens: number, info: ModelInfo): {
+    input: number;
+    output: number;
+};
 export declare function listModels(): string[];

package/dist/models.js

@@ -21,8 +21,8 @@ export const MODELS = {
     'claude-sonnet': { input: 2.00, output: 10.00, context: 1000000 },
     'claude-haiku': { input: 1.00, output: 5.00, context: 200000 },
     // xAI
-    'grok-4.20': { input: 1.25, output: 2.50, context: 1000000 },
-    'grok-4-1-fast': { input: 1.25, output: 2.50, context: 1000000 },
+    'grok-4.20': { input: 1.25, output: 2.50, context: 1000000, longContextThreshold: 200000, longContextInput: 2.50, longContextOutput: 5.00 },
+    'grok-4-1-fast': { input: 1.25, output: 2.50, context: 1000000, longContextThreshold: 200000, longContextInput: 2.50, longContextOutput: 5.00 },
     // Google
     'gemini-3.1-pro': { input: 2.00, output: 12.00, context: 1000000 },
     'gemini-2.5-flash': { input: 0.30, output: 2.50, context: 1000000 },
@@ -34,6 +34,20 @@ export const MODELS = {
 export function getModel(name) {
     return MODELS[name] || MODELS[name.toLowerCase()];
 }
+// Resolve the actual billable rate for a given token count — applies the
+// long-context tier above if the model has one and tokens cross it.
+// preflight()/preflightRoute() should go through this, not read
+// .input/.output directly, or a Grok call over 200K tokens gets silently
+// under-costed at the base rate.
+export function effectiveRate(tokens, info) {
+    if (info.longContextThreshold !== undefined && tokens > info.longContextThreshold) {
+        return {
+            input: info.longContextInput ?? info.input,
+            output: info.longContextOutput ?? info.output,
+        };
+    }
+    return { input: info.input, output: info.output };
+}
 export function listModels() {
     return Object.keys(MODELS);
 }

package/dist/preflight.js

@@ -1,14 +1,17 @@
 import { slash } from './slash.js';
-import { getModel, MODELS } from './models.js';
+import { getModel, MODELS, effectiveRate } from './models.js';
 import { PROVIDER_MODELS, providerOf } from './providers.js';
 import { shouldRoute, isModelAllowed } from './config.js';
 /**
  * Compute cost of a prompt of `tokens` tokens on the given ModelInfo.
  * Uses input-token price only (output is not known at preflight time).
- * Rounded to 6 decimals.
+ * Rounded to 6 decimals. Goes through effectiveRate() so a long-context
+ * model (currently only Grok, over 200K tokens) isn't silently costed at
+ * its base rate — fixed 2026-08-23.
  */
 function computeCost(tokens, info) {
-    return Math.round(((tokens / 1000000) * info.input) * 1000000) / 1000000;
+    const rate = effectiveRate(tokens, info);
+    return Math.round(((tokens / 1000000) * rate.input) * 1000000) / 1000000;
 }
 function buildAlternative(model, originalCost, tokens, info) {
     const altCost = computeCost(tokens, info);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "slash-tokens",
-  "version": "1.5.6",
+  "version": "1.5.7",
   "description": "Token Optimization for Context Engineers. 4.8 KB WASM. Sub-millisecond. Zero dependencies.",
   "main": "dist/index.js",
   "module": "dist/index.js",