ai-lcr 0.6.4 → 0.7.0

package/CHANGELOG.md

@@ -4,6 +4,69 @@ All notable changes to `ai-lcr` are documented here. The format follows
 [Keep a Changelog](https://keepachangelog.com/), and the project adheres to
 [Semantic Versioning](https://semver.org/).
 
+## [0.7.0] — 2026-06-20
+
+The text router now records the **provider-reported actual cost** when a provider
+returns one, instead of always estimating from the price table. The table becomes
+the routing input and the drift baseline (`estCostUsd`); the recorded `costUsd` is
+the real bill wherever the provider gives it.
+
+### Why
+
+A static price table can only encode one price per model, but an aggregator
+(OpenRouter) routes a single model across many sub-providers whose prices differ
+several-fold, picking one per call — so `tokens × table` is structurally unable to
+match the bill for multi-provider models (measured: `deepseek-v4-pro` reconciled at
+~57% of the real cost, while single-provider models like Gemini/Claude/GPT matched
+at 100%). The provider's own number already accounts for which sub-provider served,
+every token kind (cache read/write, reasoning), and fees — none of which a flat
+table can track.
+
+### Added
+
+- **`costUsd` prefers the provider-reported actual cost** (text path). Read from
+  OpenRouter's `providerMetadata.openrouter.usage` —
+  `costDetails.upstreamInferenceCost` (the real upstream / BYOK model spend) when
+  present, otherwise `cost` (the credit charge) — and from an OpenAI-compatible
+  provider's `estimated_cost` on the raw usage body. Requires the caller to enable
+  usage accounting on the provider (e.g. OpenRouter `usage: { include: true }`);
+  without it, behavior is unchanged.
+- **`estCostUsd` is now set on text records** (previously media-only) — the
+  price-table prediction for the same usage. `costUsd − estCostUsd` is the
+  price-table drift signal, so a dashboard's drift panel now works for text too.
+
+### Changed
+
+- When no provider cost is reported, `costUsd` still equals the price-table
+  estimate (and `estCostUsd` equals it, so no drift is flagged) — a pure fallback,
+  fully backward-compatible. The streaming path reads the reported cost from the
+  `finish` chunk's `providerMetadata`.
+
+## [0.6.5] — 2026-06-16
+
+Bundled price table now covers the open-weights labs, not just the Western
+proprietary makers — so `autoPrice` resolves Qwen, Kimi, MiniMax, and GLM routes
+out of the box (previously they needed a hand-typed `cost`).
+
+### Added
+
+- **`MODEL_PRICES` now includes the open-weights makers** — Qwen (Alibaba /
+  `dashscope`), Kimi (Moonshot), MiniMax, and GLM (Z.ai), alongside the existing
+  DeepSeek. 55 new first-party list prices (229 → 284 entries), keyed by each
+  maker's own bare model id (`qwen-plus`, `kimi-k2.5`, `MiniMax-M2`, `glm-4.6`,
+  …). The generator's `ALLOW` set gained `dashscope` / `moonshot` / `minimax` /
+  `zai`; no existing price changed.
+
+### Notes
+
+- These are **first-party** list rates (the maker's own API). A dedicated
+  inference *host* (DeepInfra, …) is often cheaper and uses HF-style ids
+  (`Qwen/Qwen3-…`) that won't match these bare keys — for an aggregator route,
+  keep passing an explicit `cost` or `discount`. The bundled rate is the
+  `autoPrice` baseline for the maker's own provider and a reference for the rest.
+- Aggregators (deepinfra, together, fireworks, groq, openrouter) remain
+  deliberately excluded from the table — their prices drift per-model.
+
 ## [0.6.4] — 2026-06-16
 
 DX improvements that eliminate per-project boilerplate for consumers.

package/README.md

@@ -96,7 +96,7 @@ const lcr = createLCR({
 });
 ```
 
-The same pattern works for any vendor's native SDK provider — `@ai-sdk/anthropic`, `@ai-sdk/google`, `@ai-sdk/openai`, `@ai-sdk/xai`, and so on. They all return `LanguageModelV3`, so you can mix a native vendor API with aggregators in one model's list. Native APIs are narrow (only that vendor's models) but featureful; aggregators are broad. **Official-first + aggregator-fallback** is the natural LCR shape.
+The same pattern works for any vendor's native SDK provider — `@ai-sdk/anthropic`, `@ai-sdk/google`, `@ai-sdk/openai`, `@ai-sdk/xai`, and so on. `ProviderEntry` accepts `AnyLanguageModel` — a duck-typed interface (`doGenerate` + `doStream` + `provider` + `modelId`) that any AI SDK model satisfies regardless of spec version (V2 or V3), so you never need `as`-casts at the integration boundary. Native APIs are narrow (only that vendor's models) but featureful; aggregators are broad. **Official-first + aggregator-fallback** is the natural LCR shape.
 
 ## Cheapest route for open-weights models (DeepInfra)
 
@@ -138,9 +138,50 @@ const lcr = createLCR({
 
 DeepInfra carries open weights only — no first-party Claude / GPT / Gemini. For those closed models, route through OpenRouter or a discount gateway instead.
 
+## Skip the boilerplate (`DEFAULT_PROVIDERS`)
+
+Every project that routes through OpenRouter, DeepInfra, TokenMart, DeepSeek, etc. redeclares the same `baseURL` + `apiKeyEnv` pair. `DEFAULT_PROVIDERS` is a bundled dictionary — import what you need instead of copy-pasting URLs:
+
+```ts
+import { DEFAULT_PROVIDERS } from "ai-lcr";
+import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
+
+// Pick the providers you use — type-safe, no hardcoded URLs.
+const deepinfra = createOpenAICompatible({
+  name: "deepinfra",
+  baseURL: DEFAULT_PROVIDERS.deepinfra.baseURL,
+  apiKey: process.env[DEFAULT_PROVIDERS.deepinfra.apiKeyEnv],
+});
+```
+
+Available providers:
+
+| Key | Base URL | Env var |
+|---|---|---|
+| `openrouter` | `https://openrouter.ai/api/v1` | `OPENROUTER_API_KEY` |
+| `deepinfra` | `https://api.deepinfra.com/v1/openai` | `DEEPINFRA_API_KEY` |
+| `tokenmart` | `https://model.service-inference.ai/v1` | `INFERENCE_API_KEY` |
+| `deepseek` | `https://api.deepseek.com` | `DEEPSEEK_API_KEY` |
+| `kunavo` | `https://api.kunavo.com/v1` | `KUNAVO_API_KEY` |
+| `runware` | `https://api.runware.ai/v1` | `RUNWARE_API_KEY` |
+| `fal` | `https://queue.fal.run` | `FAL_KEY` |
+
+A common pattern is to subset `DEFAULT_PROVIDERS` into a project-local type for compile-time safety:
+
+```ts
+import { DEFAULT_PROVIDERS } from "ai-lcr";
+
+type ProviderId = "deepinfra" | "openrouter";
+
+export const PROVIDERS = {
+  deepinfra: DEFAULT_PROVIDERS.deepinfra,
+  openrouter: DEFAULT_PROVIDERS.openrouter,
+} satisfies Record<ProviderId, { baseURL: string; apiKeyEnv: string }>;
+```
+
 ## Zero-config pricing (`autoPrice`)
 
-Typing `cost: { input, output }` for every provider is the tedious part. `autoPrice: true` fills any entry that has no explicit `cost` from a **bundled price table** (`MODEL_PRICES`) — official first-party rates for the native makers (OpenAI, Anthropic, Google, DeepSeek, xAI, Mistral), keyed by the bare model id you already pass to the provider:
+Typing `cost: { input, output }` for every provider is the tedious part. `autoPrice: true` fills any entry that has no explicit `cost` from a **bundled price table** (`MODEL_PRICES`) — official first-party rates for the native makers (OpenAI, Anthropic, Google, xAI, Mistral, plus the open-weights labs DeepSeek, Qwen, Kimi, MiniMax, GLM), keyed by the bare model id you already pass to the provider:
 
 ```ts
 const lcr = createLCR({
@@ -161,7 +202,7 @@ Three rules keep it predictable:
 
 - **Off by default.** Unpriced entries stay unpriced (the pre-existing behavior), so turning `autoPrice` on never silently re-prices a model — and an **explicit `cost` always wins** over the table.
 - **`discount` is the reseller knob.** A flat-% aggregator (Kunavo −20%) becomes `discount: 0.2` instead of a hand-typed number; it scales input, output, and `cacheRead` alike, and only applies when the table fills the entry. Variable-discount providers (TokenMart) still want explicit per-model `cost`.
-- **Native makers only.** The table carries first-party list prices — the cheapest, most-featureful "go direct" route. Open-weights hosts (DeepInfra) and breadth aggregators (OpenRouter) aren't in it; price those explicitly.
+- **Native makers only.** The table carries first-party list prices, keyed by each maker's own bare id (`qwen-plus`, `glm-4.6`, `kimi-k2.5`, `MiniMax-M2`). It's the autoPrice baseline when you route through that maker's own API. Open-weights *hosts* (DeepInfra uses HF-style ids like `Qwen/Qwen3-…`) and breadth aggregators (OpenRouter) aren't keyed here — price those with explicit `cost` or `discount`.
 
 Look a price up yourself with `getModelPrice("claude-sonnet-4-6")`. The table is generated from [LiteLLM's price map](https://github.com/BerriAI/litellm) (MIT) — refresh with `node scripts/gen-text-prices.mjs`.
 
@@ -315,6 +356,26 @@ const lcr = createLCR({
 });
 ```
 
+### Convention-based sink (`createEnvSink`)
+
+If your app uses the standard env vars (`LCR_INGEST_URL`, `LCR_PROJECT`, `LCR_INGEST_KEY`), you don't need to wire `createHttpSink` at all — `createEnvSink` reads them for you and returns a ready-to-use `onCall` handler (or `undefined` when `LCR_INGEST_URL` is unset, so local dev stays quiet):
+
+```ts
+import { createEnvSink } from "ai-lcr";
+import { after } from "next/server";
+
+export const lcrCallSink = createEnvSink(after);
+// → use as `onCall: lcrCallSink` in createLCR
+```
+
+The only required argument is `dispatch` — a framework-specific fire-and-forget runner (Next.js: `after`, Cloudflare: `ctx.waitUntil`, plain servers: `(fn) => fn()`). Env vars:
+
+| Var | Required | Description |
+|---|---|---|
+| `LCR_INGEST_URL` | yes (no URL → sink is `undefined`) | Dashboard origin, e.g. `https://lcr.ideamarketfit.com` |
+| `LCR_PROJECT` | no | Project tag merged into each payload; falls back to `SITE_KEY` |
+| `LCR_INGEST_KEY` | no | Bearer token (only if the dashboard sets `INGEST_KEY`) |
+
 ### The companion dashboard ([`ai-lcr-dashboard`](https://github.com/ai-lcr/ai-lcr-dashboard))
 
 <p align="center">

package/README.zh-CN.md

@@ -96,7 +96,7 @@ const lcr = createLCR({
 });
 ```
 
-同样的模式适用于任何厂商的原生 SDK provider——`@ai-sdk/anthropic`、`@ai-sdk/google`、`@ai-sdk/openai`、`@ai-sdk/xai` 等等。它们都返回 `LanguageModelV3`，所以你可以在一个模型的列表里把厂商原生 API 和聚合器混着用。原生 API 覆盖窄（只有该厂商自己的模型）但特性全；聚合器覆盖广。**官方优先 + 聚合器兜底** 正是 LCR 最自然的形态。
+同样的模式适用于任何厂商的原生 SDK provider——`@ai-sdk/anthropic`、`@ai-sdk/google`、`@ai-sdk/openai`、`@ai-sdk/xai` 等等。`ProviderEntry` 接受 `AnyLanguageModel`——一个鸭子类型接口（`doGenerate` + `doStream` + `provider` + `modelId`），任何 AI SDK model 无论 V2 还是 V3 spec 都满足，集成边界无需 `as` 强转。原生 API 覆盖窄（只有该厂商自己的模型）但特性全；聚合器覆盖广。**官方优先 + 聚合器兜底** 正是 LCR 最自然的形态。
 
 ## 开源权重模型的最便宜路由（DeepInfra）
 
@@ -138,6 +138,47 @@ const lcr = createLCR({
 
 DeepInfra 只承载开源权重——没有第一方 Claude / GPT / Gemini。那些闭源模型请走 OpenRouter 或折扣中转。
 
+## 省掉样板代码（`DEFAULT_PROVIDERS`）
+
+每个路由 OpenRouter、DeepInfra、TokenMart、DeepSeek 等的项目都要重复声明相同的 `baseURL` + `apiKeyEnv`。`DEFAULT_PROVIDERS` 是一份内置字典——import 你需要的那几个就行，不用再复制粘贴 URL：
+
+```ts
+import { DEFAULT_PROVIDERS } from "ai-lcr";
+import { createOpenAICompatible } from "@ai-sdk/openai-compatible";
+
+// 按需取——类型安全，无硬编码 URL。
+const deepinfra = createOpenAICompatible({
+  name: "deepinfra",
+  baseURL: DEFAULT_PROVIDERS.deepinfra.baseURL,
+  apiKey: process.env[DEFAULT_PROVIDERS.deepinfra.apiKeyEnv],
+});
+```
+
+可用 provider：
+
+| Key | Base URL | Env 变量 |
+|---|---|---|
+| `openrouter` | `https://openrouter.ai/api/v1` | `OPENROUTER_API_KEY` |
+| `deepinfra` | `https://api.deepinfra.com/v1/openai` | `DEEPINFRA_API_KEY` |
+| `tokenmart` | `https://model.service-inference.ai/v1` | `INFERENCE_API_KEY` |
+| `deepseek` | `https://api.deepseek.com` | `DEEPSEEK_API_KEY` |
+| `kunavo` | `https://api.kunavo.com/v1` | `KUNAVO_API_KEY` |
+| `runware` | `https://api.runware.ai/v1` | `RUNWARE_API_KEY` |
+| `fal` | `https://queue.fal.run` | `FAL_KEY` |
+
+常见用法是取 `DEFAULT_PROVIDERS` 的子集，并声明一个项目级类型保证编译安全：
+
+```ts
+import { DEFAULT_PROVIDERS } from "ai-lcr";
+
+type ProviderId = "deepinfra" | "openrouter";
+
+export const PROVIDERS = {
+  deepinfra: DEFAULT_PROVIDERS.deepinfra,
+  openrouter: DEFAULT_PROVIDERS.openrouter,
+} satisfies Record<ProviderId, { baseURL: string; apiKeyEnv: string }>;
+```
+
 ## 它如何路由
 
 1. **最便宜优先。** provider 按顺序依次尝试——把它们排成最便宜优先，或设置 `autoSort: true` 让它按 `cost` 自动排序。
@@ -221,7 +262,17 @@ interface CallRecord {
 
 **节约怎么算才诚实：** `baselineKind` 说明 `baselineUsd` 是哪种基线——文本是**链尾兜底 provider 的列表价**（`"last-leg"`，故意不取最贵的一条：prompt 缓存可能让标价更便宜的那家在缓存重的调用上反而更贵，取最大值会凭空造出"节约"）；媒体是**模型厂商官方第一方价**（`"official"`，按实际秒数算），查不到官方价时退化为你配置里最贵的路由（`"priciest-route"`，自我参照，仅说明跨 provider 价差）。
 
-**送进收集器：** `createHttpSink` 把每条记录 POST 到任意 endpoint（serverless 上传 Next.js 的 `after` 作 `dispatch` 防止被掐断）。配套的自托管 dashboard [`ai-lcr-dashboard`](https://github.com/ai-lcr/ai-lcr-dashboard)（Next.js + Postgres，Vercel 一键部署）专为这些记录而建：花费 vs 节约趋势、各 provider failover 健康度、媒体 $/秒 与 $/张、以及**价格漂移面板**——某条 model@provider 路由的实报账单与价格表偏差超过 ±20% 时点名示警（约 100× 基本就是美元当美分的笔误）。只存元数据，不存 prompt 和输出。
+**送进收集器：** `createHttpSink` 把每条记录 POST 到任意 endpoint（serverless 上传 Next.js 的 `after` 作 `dispatch` 防止被掐断）。如果你用标准环境变量（`LCR_INGEST_URL`、`LCR_PROJECT`、`LCR_INGEST_KEY`），`createEnvSink` 全部替你读好——三行搞定：
+
+```ts
+import { createEnvSink } from "ai-lcr";
+import { after } from "next/server";
+export const lcrCallSink = createEnvSink(after);
+```
+
+`LCR_INGEST_URL` 不设 → sink 是 `undefined`，本地开发自动静默。唯一必传参数是 `dispatch`——框架相关的 fire-and-forget runner（Next.js: `after`；Cloudflare: `ctx.waitUntil`；长驻服务: `(fn) => fn()`）。
+
+配套的自托管 dashboard [`ai-lcr-dashboard`](https://github.com/ai-lcr/ai-lcr-dashboard)（Next.js + Postgres，Vercel 一键部署）专为这些记录而建：花费 vs 节约趋势、各 provider failover 健康度、媒体 $/秒 与 $/张、以及**价格漂移面板**——某条 model@provider 路由的实报账单与价格表偏差超过 ±20% 时点名示警（约 100× 基本就是美元当美分的笔误）。只存元数据，不存 prompt 和输出。
 
 ## 支持的 provider
 

package/dist/index.cjs

@@ -341,6 +341,20 @@ function cacheSavingForUsage(cost, inputTokens, cacheReadTokens) {
   const cached = Math.min(Math.max(cacheReadTokens, 0), inputTokens);
   return cached / 1e6 * (cost.input - cost.cacheRead);
 }
+function reportedCost(providerMetadata, usage) {
+  const orUsage = providerMetadata?.openrouter?.usage;
+  if (orUsage) {
+    const upstream = orUsage.costDetails?.upstreamInferenceCost;
+    if (typeof upstream === "number" && upstream > 0) return upstream;
+    if (typeof orUsage.cost === "number") return orUsage.cost;
+  }
+  const raw = usage?.raw;
+  if (raw) {
+    const est = raw["estimated_cost"] ?? raw["cost"];
+    if (typeof est === "number") return est;
+  }
+  return void 0;
+}
 function requestIdFrom(options) {
   const raw = options.providerOptions?.lcr?.requestId;
   return typeof raw === "string" && raw.length > 0 ? raw : void 0;
@@ -539,12 +553,13 @@ var LcrFallbackModel = class {
     return baseline;
   }
   /** Winner settled: record the attempt, fire `onCost` (compat) + `onCall`. */
-  finalizeOk(ctx, provider, attemptStart, usage, ttftMs) {
+  finalizeOk(ctx, provider, attemptStart, usage, ttftMs, providerMetadata) {
     ctx.attempts.push({ provider: provider.label, ok: true, latencyMs: Date.now() - attemptStart });
     const inputTokens = usage?.inputTokens?.total ?? 0;
     const outputTokens = usage?.outputTokens?.total ?? 0;
     const cacheReadTokens = usage?.inputTokens?.cacheRead ?? 0;
-    const costUsd = provider.cost ? costForUsage(provider.cost, inputTokens, outputTokens, cacheReadTokens) : 0;
+    const estCostUsd = provider.cost ? costForUsage(provider.cost, inputTokens, outputTokens, cacheReadTokens) : void 0;
+    const costUsd = reportedCost(providerMetadata, usage) ?? estCostUsd ?? 0;
     const cachedSavingUsd = provider.cost ? cacheSavingForUsage(provider.cost, inputTokens, cacheReadTokens) : 0;
     const usageMissing = inputTokens === 0 && outputTokens === 0;
     const emptyCompletion = inputTokens > 0 && outputTokens === 0;
@@ -579,6 +594,7 @@ var LcrFallbackModel = class {
       outputTokens,
       ...cacheReadTokens > 0 ? { cachedInputTokens: cacheReadTokens } : {},
       costUsd,
+      ...estCostUsd !== void 0 ? { estCostUsd } : {},
       ...baselineUsd !== void 0 ? { baselineUsd, baselineKind: "last-leg" } : {},
       ...cachedSavingUsd > 0 ? { cachedSavingUsd } : {},
       ...ctx.requestId ? { requestId: ctx.requestId } : {},
@@ -635,7 +651,7 @@ var LcrFallbackModel = class {
         }
         this.recordProviderSuccess(idx);
         this.settleSticky(idx);
-        this.finalizeOk(ctx, provider, attemptStart, result.usage);
+        this.finalizeOk(ctx, provider, attemptStart, result.usage, void 0, result.providerMetadata);
         if (cache && cacheKey !== void 0 && ctx.settled?.cacheable) {
           this.storeCache(cacheKey, { kind: "generate", result, meta: ctx.settled.meta });
         }
@@ -767,6 +783,7 @@ var LcrFallbackModel = class {
     const servingIdx = idx;
     const servingPos = p;
     let usage;
+    let finishProviderMetadata;
     let contentStreamed = false;
     let ttftMs;
     const stream = new ReadableStream({
@@ -783,6 +800,7 @@ var LcrFallbackModel = class {
             if (done) break;
             if (value.type === "finish") {
               usage = value.usage;
+              finishProviderMetadata = value.providerMetadata;
               const out = value.usage?.outputTokens?.total ?? 0;
               const inp = value.usage?.inputTokens?.total ?? 0;
               if (inp > 0 && out === 0 && !contentStreamed && servingPos + 1 < n) {
@@ -797,7 +815,7 @@ var LcrFallbackModel = class {
           }
           self.recordProviderSuccess(servingIdx);
           self.settleSticky(servingIdx);
-          self.finalizeOk(ctx, servingProvider, servingAttemptStart, usage, ttftMs);
+          self.finalizeOk(ctx, servingProvider, servingAttemptStart, usage, ttftMs, finishProviderMetadata);
           controller.close();
         } catch (error) {
           self.emitError(error, servingProvider.label);
@@ -1003,6 +1021,16 @@ var MODEL_PRICES = {
   "gemini-gemma-2-9b-it": { input: 0.35, output: 1.05 },
   "gemini-pro-latest": { input: 1.25, output: 10, cacheRead: 0.125 },
   "gemini-robotics-er-1.5-preview": { input: 0.3, output: 2.5 },
+  "glm-4-32b-0414-128k": { input: 0.1, output: 0.1 },
+  "glm-4.5": { input: 0.6, output: 2.2 },
+  "glm-4.5-air": { input: 0.2, output: 1.1 },
+  "glm-4.5-airx": { input: 1.1, output: 4.5 },
+  "glm-4.5-x": { input: 2.2, output: 8.9 },
+  "glm-4.5v": { input: 0.6, output: 1.8 },
+  "glm-4.6": { input: 0.6, output: 2.2, cacheRead: 0.11 },
+  "glm-4.7": { input: 0.6, output: 2.2, cacheRead: 0.11 },
+  "glm-5": { input: 1, output: 3.2, cacheRead: 0.2 },
+  "glm-5-code": { input: 1.2, output: 5, cacheRead: 0.3 },
   "gpt-3.5-turbo": { input: 0.5, output: 1.5 },
   "gpt-3.5-turbo-0125": { input: 0.5, output: 1.5 },
   "gpt-3.5-turbo-1106": { input: 1, output: 2 },
@@ -1117,6 +1145,18 @@ var MODEL_PRICES = {
   "grok-code-fast-1": { input: 0.2, output: 1.5, cacheRead: 0.02 },
   "grok-code-fast-1-0825": { input: 0.2, output: 1.5, cacheRead: 0.02 },
   "grok-vision-beta": { input: 5, output: 15 },
+  "kimi-k2-0711-preview": { input: 0.6, output: 2.5, cacheRead: 0.15 },
+  "kimi-k2-0905-preview": { input: 0.6, output: 2.5, cacheRead: 0.15 },
+  "kimi-k2-thinking": { input: 0.6, output: 2.5, cacheRead: 0.15 },
+  "kimi-k2-thinking-turbo": { input: 1.15, output: 8, cacheRead: 0.15 },
+  "kimi-k2-turbo-preview": { input: 1.15, output: 8, cacheRead: 0.15 },
+  "kimi-k2.5": { input: 0.6, output: 3, cacheRead: 0.1 },
+  "kimi-k2.6": { input: 0.95, output: 4, cacheRead: 0.16 },
+  "kimi-latest": { input: 2, output: 5, cacheRead: 0.15 },
+  "kimi-latest-128k": { input: 2, output: 5, cacheRead: 0.15 },
+  "kimi-latest-32k": { input: 1, output: 3, cacheRead: 0.15 },
+  "kimi-latest-8k": { input: 0.2, output: 2, cacheRead: 0.15 },
+  "kimi-thinking-preview": { input: 0.6, output: 2.5, cacheRead: 0.15 },
   "labs-devstral-small-2512": { input: 0.1, output: 0.3 },
   "magistral-medium-1-2-2509": { input: 2, output: 5 },
   "magistral-medium-2506": { input: 2, output: 5 },
@@ -1125,6 +1165,12 @@ var MODEL_PRICES = {
   "magistral-small-1-2-2509": { input: 0.5, output: 1.5 },
   "magistral-small-2506": { input: 0.5, output: 1.5 },
   "magistral-small-latest": { input: 0.5, output: 1.5 },
+  "MiniMax-M2": { input: 0.3, output: 1.2, cacheRead: 0.03 },
+  "MiniMax-M2.1": { input: 0.3, output: 1.2, cacheRead: 0.03 },
+  "MiniMax-M2.1-lightning": { input: 0.3, output: 2.4, cacheRead: 0.03 },
+  "MiniMax-M2.5": { input: 0.3, output: 1.2, cacheRead: 0.03 },
+  "MiniMax-M2.5-lightning": { input: 0.3, output: 2.4, cacheRead: 0.03 },
+  "MiniMax-M3": { input: 0.6, output: 2.4, cacheRead: 0.12 },
   "ministral-3-14b-2512": { input: 0.2, output: 0.2 },
   "ministral-3-3b-2512": { input: 0.1, output: 0.1 },
   "ministral-3-8b-2512": { input: 0.15, output: 0.15 },
@@ -1145,6 +1191,16 @@ var MODEL_PRICES = {
   "mistral-small-3-2-2506": { input: 0.06, output: 0.18 },
   "mistral-small-latest": { input: 0.06, output: 0.18 },
   "mistral-tiny": { input: 0.25, output: 0.25 },
+  "moonshot-v1-128k": { input: 2, output: 5 },
+  "moonshot-v1-128k-0430": { input: 2, output: 5 },
+  "moonshot-v1-128k-vision-preview": { input: 2, output: 5 },
+  "moonshot-v1-32k": { input: 1, output: 3 },
+  "moonshot-v1-32k-0430": { input: 1, output: 3 },
+  "moonshot-v1-32k-vision-preview": { input: 1, output: 3 },
+  "moonshot-v1-8k": { input: 0.2, output: 2 },
+  "moonshot-v1-8k-0430": { input: 0.2, output: 2 },
+  "moonshot-v1-8k-vision-preview": { input: 0.2, output: 2 },
+  "moonshot-v1-auto": { input: 2, output: 5 },
   "o1": { input: 15, output: 60, cacheRead: 7.5 },
   "o1-2024-12-17": { input: 15, output: 60, cacheRead: 7.5 },
   "o3": { input: 2, output: 8, cacheRead: 0.5 },
@@ -1161,7 +1217,24 @@ var MODEL_PRICES = {
   "open-mixtral-8x7b": { input: 0.7, output: 0.7 },
   "pixtral-12b-2409": { input: 0.15, output: 0.15 },
   "pixtral-large-2411": { input: 2, output: 6 },
-  "pixtral-large-latest": { input: 2, output: 6 }
+  "pixtral-large-latest": { input: 2, output: 6 },
+  "qwen-coder": { input: 0.3, output: 1.5 },
+  "qwen-max": { input: 1.6, output: 6.4 },
+  "qwen-plus": { input: 0.4, output: 1.2 },
+  "qwen-plus-2025-01-25": { input: 0.4, output: 1.2 },
+  "qwen-plus-2025-04-28": { input: 0.4, output: 1.2 },
+  "qwen-plus-2025-07-14": { input: 0.4, output: 1.2 },
+  "qwen-turbo": { input: 0.05, output: 0.2 },
+  "qwen-turbo-2024-11-01": { input: 0.05, output: 0.2 },
+  "qwen-turbo-2025-04-28": { input: 0.05, output: 0.2 },
+  "qwen-turbo-latest": { input: 0.05, output: 0.2 },
+  "qwen3-next-80b-a3b-instruct": { input: 0.15, output: 1.2 },
+  "qwen3-next-80b-a3b-thinking": { input: 0.15, output: 1.2 },
+  "qwen3-vl-235b-a22b-instruct": { input: 0.4, output: 1.6 },
+  "qwen3-vl-235b-a22b-thinking": { input: 0.4, output: 4 },
+  "qwen3-vl-32b-instruct": { input: 0.16, output: 0.64 },
+  "qwen3-vl-32b-thinking": { input: 0.16, output: 2.87 },
+  "qwq-plus": { input: 0.8, output: 2.4 }
 };
 
 // src/media-official.ts

package/dist/index.js

@@ -287,6 +287,20 @@ function cacheSavingForUsage(cost, inputTokens, cacheReadTokens) {
   const cached = Math.min(Math.max(cacheReadTokens, 0), inputTokens);
   return cached / 1e6 * (cost.input - cost.cacheRead);
 }
+function reportedCost(providerMetadata, usage) {
+  const orUsage = providerMetadata?.openrouter?.usage;
+  if (orUsage) {
+    const upstream = orUsage.costDetails?.upstreamInferenceCost;
+    if (typeof upstream === "number" && upstream > 0) return upstream;
+    if (typeof orUsage.cost === "number") return orUsage.cost;
+  }
+  const raw = usage?.raw;
+  if (raw) {
+    const est = raw["estimated_cost"] ?? raw["cost"];
+    if (typeof est === "number") return est;
+  }
+  return void 0;
+}
 function requestIdFrom(options) {
   const raw = options.providerOptions?.lcr?.requestId;
   return typeof raw === "string" && raw.length > 0 ? raw : void 0;
@@ -485,12 +499,13 @@ var LcrFallbackModel = class {
     return baseline;
   }
   /** Winner settled: record the attempt, fire `onCost` (compat) + `onCall`. */
-  finalizeOk(ctx, provider, attemptStart, usage, ttftMs) {
+  finalizeOk(ctx, provider, attemptStart, usage, ttftMs, providerMetadata) {
     ctx.attempts.push({ provider: provider.label, ok: true, latencyMs: Date.now() - attemptStart });
     const inputTokens = usage?.inputTokens?.total ?? 0;
     const outputTokens = usage?.outputTokens?.total ?? 0;
     const cacheReadTokens = usage?.inputTokens?.cacheRead ?? 0;
-    const costUsd = provider.cost ? costForUsage(provider.cost, inputTokens, outputTokens, cacheReadTokens) : 0;
+    const estCostUsd = provider.cost ? costForUsage(provider.cost, inputTokens, outputTokens, cacheReadTokens) : void 0;
+    const costUsd = reportedCost(providerMetadata, usage) ?? estCostUsd ?? 0;
     const cachedSavingUsd = provider.cost ? cacheSavingForUsage(provider.cost, inputTokens, cacheReadTokens) : 0;
     const usageMissing = inputTokens === 0 && outputTokens === 0;
     const emptyCompletion = inputTokens > 0 && outputTokens === 0;
@@ -525,6 +540,7 @@ var LcrFallbackModel = class {
       outputTokens,
       ...cacheReadTokens > 0 ? { cachedInputTokens: cacheReadTokens } : {},
       costUsd,
+      ...estCostUsd !== void 0 ? { estCostUsd } : {},
       ...baselineUsd !== void 0 ? { baselineUsd, baselineKind: "last-leg" } : {},
       ...cachedSavingUsd > 0 ? { cachedSavingUsd } : {},
       ...ctx.requestId ? { requestId: ctx.requestId } : {},
@@ -581,7 +597,7 @@ var LcrFallbackModel = class {
         }
         this.recordProviderSuccess(idx);
         this.settleSticky(idx);
-        this.finalizeOk(ctx, provider, attemptStart, result.usage);
+        this.finalizeOk(ctx, provider, attemptStart, result.usage, void 0, result.providerMetadata);
         if (cache && cacheKey !== void 0 && ctx.settled?.cacheable) {
           this.storeCache(cacheKey, { kind: "generate", result, meta: ctx.settled.meta });
         }
@@ -713,6 +729,7 @@ var LcrFallbackModel = class {
     const servingIdx = idx;
     const servingPos = p;
     let usage;
+    let finishProviderMetadata;
     let contentStreamed = false;
     let ttftMs;
     const stream = new ReadableStream({
@@ -729,6 +746,7 @@ var LcrFallbackModel = class {
             if (done) break;
             if (value.type === "finish") {
               usage = value.usage;
+              finishProviderMetadata = value.providerMetadata;
               const out = value.usage?.outputTokens?.total ?? 0;
               const inp = value.usage?.inputTokens?.total ?? 0;
               if (inp > 0 && out === 0 && !contentStreamed && servingPos + 1 < n) {
@@ -743,7 +761,7 @@ var LcrFallbackModel = class {
           }
           self.recordProviderSuccess(servingIdx);
           self.settleSticky(servingIdx);
-          self.finalizeOk(ctx, servingProvider, servingAttemptStart, usage, ttftMs);
+          self.finalizeOk(ctx, servingProvider, servingAttemptStart, usage, ttftMs, finishProviderMetadata);
           controller.close();
         } catch (error) {
           self.emitError(error, servingProvider.label);
@@ -949,6 +967,16 @@ var MODEL_PRICES = {
   "gemini-gemma-2-9b-it": { input: 0.35, output: 1.05 },
   "gemini-pro-latest": { input: 1.25, output: 10, cacheRead: 0.125 },
   "gemini-robotics-er-1.5-preview": { input: 0.3, output: 2.5 },
+  "glm-4-32b-0414-128k": { input: 0.1, output: 0.1 },
+  "glm-4.5": { input: 0.6, output: 2.2 },
+  "glm-4.5-air": { input: 0.2, output: 1.1 },
+  "glm-4.5-airx": { input: 1.1, output: 4.5 },
+  "glm-4.5-x": { input: 2.2, output: 8.9 },
+  "glm-4.5v": { input: 0.6, output: 1.8 },
+  "glm-4.6": { input: 0.6, output: 2.2, cacheRead: 0.11 },
+  "glm-4.7": { input: 0.6, output: 2.2, cacheRead: 0.11 },
+  "glm-5": { input: 1, output: 3.2, cacheRead: 0.2 },
+  "glm-5-code": { input: 1.2, output: 5, cacheRead: 0.3 },
   "gpt-3.5-turbo": { input: 0.5, output: 1.5 },
   "gpt-3.5-turbo-0125": { input: 0.5, output: 1.5 },
   "gpt-3.5-turbo-1106": { input: 1, output: 2 },
@@ -1063,6 +1091,18 @@ var MODEL_PRICES = {
   "grok-code-fast-1": { input: 0.2, output: 1.5, cacheRead: 0.02 },
   "grok-code-fast-1-0825": { input: 0.2, output: 1.5, cacheRead: 0.02 },
   "grok-vision-beta": { input: 5, output: 15 },
+  "kimi-k2-0711-preview": { input: 0.6, output: 2.5, cacheRead: 0.15 },
+  "kimi-k2-0905-preview": { input: 0.6, output: 2.5, cacheRead: 0.15 },
+  "kimi-k2-thinking": { input: 0.6, output: 2.5, cacheRead: 0.15 },
+  "kimi-k2-thinking-turbo": { input: 1.15, output: 8, cacheRead: 0.15 },
+  "kimi-k2-turbo-preview": { input: 1.15, output: 8, cacheRead: 0.15 },
+  "kimi-k2.5": { input: 0.6, output: 3, cacheRead: 0.1 },
+  "kimi-k2.6": { input: 0.95, output: 4, cacheRead: 0.16 },
+  "kimi-latest": { input: 2, output: 5, cacheRead: 0.15 },
+  "kimi-latest-128k": { input: 2, output: 5, cacheRead: 0.15 },
+  "kimi-latest-32k": { input: 1, output: 3, cacheRead: 0.15 },
+  "kimi-latest-8k": { input: 0.2, output: 2, cacheRead: 0.15 },
+  "kimi-thinking-preview": { input: 0.6, output: 2.5, cacheRead: 0.15 },
   "labs-devstral-small-2512": { input: 0.1, output: 0.3 },
   "magistral-medium-1-2-2509": { input: 2, output: 5 },
   "magistral-medium-2506": { input: 2, output: 5 },
@@ -1071,6 +1111,12 @@ var MODEL_PRICES = {
   "magistral-small-1-2-2509": { input: 0.5, output: 1.5 },
   "magistral-small-2506": { input: 0.5, output: 1.5 },
   "magistral-small-latest": { input: 0.5, output: 1.5 },
+  "MiniMax-M2": { input: 0.3, output: 1.2, cacheRead: 0.03 },
+  "MiniMax-M2.1": { input: 0.3, output: 1.2, cacheRead: 0.03 },
+  "MiniMax-M2.1-lightning": { input: 0.3, output: 2.4, cacheRead: 0.03 },
+  "MiniMax-M2.5": { input: 0.3, output: 1.2, cacheRead: 0.03 },
+  "MiniMax-M2.5-lightning": { input: 0.3, output: 2.4, cacheRead: 0.03 },
+  "MiniMax-M3": { input: 0.6, output: 2.4, cacheRead: 0.12 },
   "ministral-3-14b-2512": { input: 0.2, output: 0.2 },
   "ministral-3-3b-2512": { input: 0.1, output: 0.1 },
   "ministral-3-8b-2512": { input: 0.15, output: 0.15 },
@@ -1091,6 +1137,16 @@ var MODEL_PRICES = {
   "mistral-small-3-2-2506": { input: 0.06, output: 0.18 },
   "mistral-small-latest": { input: 0.06, output: 0.18 },
   "mistral-tiny": { input: 0.25, output: 0.25 },
+  "moonshot-v1-128k": { input: 2, output: 5 },
+  "moonshot-v1-128k-0430": { input: 2, output: 5 },
+  "moonshot-v1-128k-vision-preview": { input: 2, output: 5 },
+  "moonshot-v1-32k": { input: 1, output: 3 },
+  "moonshot-v1-32k-0430": { input: 1, output: 3 },
+  "moonshot-v1-32k-vision-preview": { input: 1, output: 3 },
+  "moonshot-v1-8k": { input: 0.2, output: 2 },
+  "moonshot-v1-8k-0430": { input: 0.2, output: 2 },
+  "moonshot-v1-8k-vision-preview": { input: 0.2, output: 2 },
+  "moonshot-v1-auto": { input: 2, output: 5 },
   "o1": { input: 15, output: 60, cacheRead: 7.5 },
   "o1-2024-12-17": { input: 15, output: 60, cacheRead: 7.5 },
   "o3": { input: 2, output: 8, cacheRead: 0.5 },
@@ -1107,7 +1163,24 @@ var MODEL_PRICES = {
   "open-mixtral-8x7b": { input: 0.7, output: 0.7 },
   "pixtral-12b-2409": { input: 0.15, output: 0.15 },
   "pixtral-large-2411": { input: 2, output: 6 },
-  "pixtral-large-latest": { input: 2, output: 6 }
+  "pixtral-large-latest": { input: 2, output: 6 },
+  "qwen-coder": { input: 0.3, output: 1.5 },
+  "qwen-max": { input: 1.6, output: 6.4 },
+  "qwen-plus": { input: 0.4, output: 1.2 },
+  "qwen-plus-2025-01-25": { input: 0.4, output: 1.2 },
+  "qwen-plus-2025-04-28": { input: 0.4, output: 1.2 },
+  "qwen-plus-2025-07-14": { input: 0.4, output: 1.2 },
+  "qwen-turbo": { input: 0.05, output: 0.2 },
+  "qwen-turbo-2024-11-01": { input: 0.05, output: 0.2 },
+  "qwen-turbo-2025-04-28": { input: 0.05, output: 0.2 },
+  "qwen-turbo-latest": { input: 0.05, output: 0.2 },
+  "qwen3-next-80b-a3b-instruct": { input: 0.15, output: 1.2 },
+  "qwen3-next-80b-a3b-thinking": { input: 0.15, output: 1.2 },
+  "qwen3-vl-235b-a22b-instruct": { input: 0.4, output: 1.6 },
+  "qwen3-vl-235b-a22b-thinking": { input: 0.4, output: 4 },
+  "qwen3-vl-32b-instruct": { input: 0.16, output: 0.64 },
+  "qwen3-vl-32b-thinking": { input: 0.16, output: 2.87 },
+  "qwq-plus": { input: 0.8, output: 2.4 }
 };
 
 // src/media-official.ts

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ai-lcr",
-  "version": "0.6.4",
+  "version": "0.7.0",
   "description": "Least Cost Routing for LLMs — route every model call to the cheapest available provider, fall back automatically, and track real cost. Built for the Vercel AI SDK.",
   "keywords": [
     "ai",