Skip to content

7.4 billion tokens per month. 34 free LLM providers. 635 free model endpoints. All behind one /v1 endpoint, plus any custom OpenAI-compatible endpoint. Smart routing, automatic failover, encrypted keys. Personal experimentation only.

TypeScript

fix(providers): retire the OpenCode free roster and stop calling Reka free (#1202, #1249) (#1265)

Tashfeen46de66184ffe
diff --git a/client/src/components/keys/shared.tsx b/client/src/components/keys/shared.tsx
index ae7cffaf..1827fc3a 100644
--- a/client/src/components/keys/shared.tsx
+++ b/client/src/components/keys/shared.tsx
@@ -57,9 +57,9 @@ export const PLATFORMS: { value: Platform; label: string; url: string; keyless?:
{ value: 'ovh', label: 'OVH AI Endpoints (no key needed)', url: 'https://endpoints.ai.cloud.ovh.net', keyless: true },
{ value: 'llm7', label: 'LLM7 (anon ok)', url: 'https://llm7.io' },
{ value: 'huggingface', label: 'HuggingFace Router', url: 'https://huggingface.co/settings/tokens' },
- { value: 'opencode', label: 'OpenCode Zen (free key)', url: 'https://opencode.ai/auth' },
+ { value: 'opencode', label: 'OpenCode Zen (paid models only)', url: 'https://opencode.ai/auth' },
{ value: 'agnes', label: 'Agnes AI (free key)', url: 'https://platform.agnes-ai.com' },
- { value: 'reka', label: 'Reka (free key)', url: 'https://platform.reka.ai' },
+ { value: 'reka', label: 'Reka (prepaid credits)', url: 'https://platform.reka.ai' },
{ value: 'siliconflow', label: 'SiliconFlow (image + TTS)', url: 'https://siliconflow.com' },
{ value: 'routeway', label: 'Routeway (free key)', url: 'https://routeway.ai' },
{ value: 'bazaarlink', label: 'BazaarLink (free key)', url: 'https://bazaarlink.ai' },
diff --git a/docs/en/providers/01-supported-platforms.md b/docs/en/providers/01-supported-platforms.md
index 45a4ec4a..353765f4 100644
--- a/docs/en/providers/01-supported-platforms.md
+++ b/docs/en/providers/01-supported-platforms.md
@@ -36,10 +36,10 @@ Of the 43 built-in platforms, **10 use dedicated adapters** and **33 ride `OpenA
| `pollinations` | Pollinations | Keyed | Native (`PollinationsProvider`) | `GET /v1/models` answers 200 even for revoked keys, so validation probes the authenticated `/account/key` instead (#608). |
| `llm7` | LLM7.io | Keyed (anonymous works for basic models) | OpenAI-compat | 100 req/hr free tier. |
| `huggingface` | Hugging Face Router | Keyed | OpenAI-compat | `router.huggingface.co` meta-router (re-added V13); recurring $0.10/mo router credit on the free tier. |
-| `opencode` | OpenCode Zen | Keyed | OpenAI-compat | Trial-only promotional free roster, docs-confirmed ids seeded via `migrateModelsV18`. |
+| `opencode` | OpenCode Zen | Keyed | OpenAI-compat | Free roster is locked to the OpenCode client since 2026-09 (403 FreeTierError, #1249) and disabled in the catalog; paid models only. |
| `ovh` | OVHcloud AI Endpoints | Keyless | OpenAI-compat | Anonymous tier: 2 req/min per IP per model (observed stricter); authenticated tier requires a Public Cloud project with payment method on file (`migrateModelsV26`). |
| `agnes` | Agnes AI | Keyed | OpenAI-compat | Proprietary models served at $0/token promotionally; ~30 concurrent requests before 429s; 60s timeout for reasoning TTFB. |
-| `reka` | Reka | Keyed | OpenAI-compat | Free via recurring monthly credit grant (no card); balance dashboard-only. |
+| `reka` | Reka | Keyed | OpenAI-compat | No free tier for new accounts since 2026-09: prepaid credits required (#1202). Accounts that still hold credit keep working; balance dashboard-only. |
| `siliconflow` | SiliconFlow | Keyed | OpenAI-compat | Registered mainly for FREE generative-media models (FLUX.1-schnell image, CosyVoice2 TTS) routed via `services/media.ts`. |
| `routeway` | Routeway | Keyed | OpenAI-compat | Requires browser-style User-Agent (Cloudflare rejects others with error 1010); free pool observed stricter (~5 rpm) than the documented 20 rpm / 200 rpd. |
| `bazaarlink` | BazaarLink | Keyed | OpenAI-compat | Only the `auto:free` route is cataloged — direct model ids are paid (#385). |
diff --git a/docs/zh-cn/providers/01-supported-platforms.md b/docs/zh-cn/providers/01-supported-platforms.md
index d8518cf4..c08270c6 100644
--- a/docs/zh-cn/providers/01-supported-platforms.md
+++ b/docs/zh-cn/providers/01-supported-platforms.md
@@ -32,10 +32,10 @@
| `pollinations` | Pollinations | 带密钥 | 原生(`PollinationsProvider`) | 即使密钥已被撤销,`GET /v1/models` 也返回 200,因此校验改为探测需鉴权的 `/account/key`(#608)。 |
| `llm7` | LLM7.io | 带密钥(基础模型可匿名) | OpenAI 兼容 | 免费档每小时 100 次请求。 |
| `huggingface` | Hugging Face Router | 带密钥 | OpenAI 兼容 | `router.huggingface.co` 元路由器(V13 重新加入);免费档每月循环发放 $0.10 路由额度。 |
-| `opencode` | OpenCode Zen | 带密钥 | OpenAI 兼容 | 仅限试用的促销免费名册,经文档确认的 id 通过 `migrateModelsV18` 播种。 |
+| `opencode` | OpenCode Zen | 带密钥 | OpenAI 兼容 | 自 2026-09 起免费名册仅限 OpenCode 客户端使用(403 FreeTierError,#1249),已在目录中停用;仅支持付费模型。 |
| `ovh` | OVHcloud AI Endpoints | 免密钥 | OpenAI 兼容 | 匿名档:每 IP 每模型每分钟 2 次请求(实测更严);鉴权档要求绑定了支付方式的 Public Cloud 项目(`migrateModelsV26`)。 |
| `agnes` | Agnes AI | 带密钥 | OpenAI 兼容 | 专有模型以 $0/词元促销提供;约 30 个并发请求后开始出现 429;为推理首字节设 60s 超时。 |
-| `reka` | Reka | 带密钥 | OpenAI 兼容 | 通过每月循环的额度赠金免费(无需信用卡);余额只在仪表盘可见。 |
+| `reka` | Reka | 带密钥 | OpenAI 兼容 | 自 2026-09 起新账号不再免费:需预付额度(#1202)。仍有余额的账号可继续使用;余额只在仪表盘可见。 |
| `siliconflow` | SiliconFlow | 带密钥 | OpenAI 兼容 | 主要为 FREE 生成媒体模型而注册(FLUX.1-schnell 图像、CosyVoice2 TTS),经由 `services/media.ts` 路由。 |
| `routeway` | Routeway | 带密钥 | OpenAI 兼容 | 要求浏览器风格的 User-Agent(Cloudflare 对其他值报错误 1010);实测免费池约 5 rpm,比文档写的 20 rpm / 200 rpd 更严。 |
| `bazaarlink` | BazaarLink | 带密钥 | OpenAI 兼容 | 只有 `auto:free` 路由进了目录——直接指定模型 id 是付费的(#385)。 |
diff --git a/server/src/__tests__/providers/openai-compat.test.ts b/server/src/__tests__/providers/openai-compat.test.ts
index 9a974423..c0e5a8c1 100644
--- a/server/src/__tests__/providers/openai-compat.test.ts
+++ b/server/src/__tests__/providers/openai-compat.test.ts
@@ -70,36 +70,6 @@ describe('OpenAICompatProvider', () => {
expect(capturedBody.messages[0].role).toBe('user');
});
- it('evaluates function-valued extraHeaders once per request', async () => {
- let issued = 0;
- const dynamic = new OpenAICompatProvider({
- platform: 'opencode',
- name: 'Zen',
- baseUrl: 'https://api.test.com/v1',
- extraHeaders: () => ({ 'x-opencode-session': `session-${++issued}` }),
- });
- const seen: string[] = [];
- vi.spyOn(global, 'fetch').mockImplementation(async (_url, init) => {
- seen.push((init as any).headers['x-opencode-session']);
- return {
- ok: true,
- json: () => Promise.resolve({
- id: 'test-id',
- object: 'chat.completion',
- created: 123,
- model: 'test-model',
- choices: [{ index: 0, message: { role: 'assistant', content: 'hi' }, finish_reason: 'stop' }],
- usage: { prompt_tokens: 1, completion_tokens: 1, total_tokens: 2 },
- }),
- } as any;
- });
-
- await dynamic.chatCompletion('my-key', [{ role: 'user', content: 'one' }], 'test-model');
- await dynamic.chatCompletion('my-key', [{ role: 'user', content: 'two' }], 'test-model');
-
- expect(seen).toEqual(['session-1', 'session-2']);
- });
-
describe('Moonshot assistant `partial` prefill flag (#1038)', () => {
beforeEach(() => {
// The custom-platform SSRF guard resolves public hosts before every
diff --git a/server/src/providers/index.ts b/server/src/providers/index.ts
index bc52e4e2..057f51b6 100644
--- a/server/src/providers/index.ts
+++ b/server/src/providers/index.ts
@@ -1,5 +1,4 @@
import type { Platform } from '@freellmapi/shared/types.js';
-import { randomUUID } from 'node:crypto';
import type { BaseProvider } from './base.js';
import { GoogleProvider } from './google.js';
import { OpenAICompatProvider } from './openai-compat.js';
@@ -258,51 +257,15 @@ register(new OpenAICompatProvider({
// (no card required — billing only applies to paid models). The free roster is
// trial-only and prompts/outputs may be used to improve the models, so we seed
// just the docs-confirmed free IDs (migrateModelsV18) with conservative limits.
-// The Console gateway additionally expects the request to look like it came
-// from the OpenCode client: without x-opencode-session it answers
-// MissingSessionID, and the free roster answers "OpenCode's free tier can only
-// be used in OpenCode" even with a valid key. Fresh session/request ids per
-// call mirror the CLI (both the free roster and the paid Go plan were verified
-// against this header set).
-// The UA advertises the newest published OpenCode release so an upstream
-// version check cannot age into a block. The npm registry (the artifact the
-// CLI itself ships from) is the cheapest source; a failed lookup keeps the
-// last known version and the built-in fallback covers a cold first boot.
-const OPENCODE_NPM_LATEST = 'https://registry.npmjs.org/opencode-ai/latest';
-const OPENCODE_UA_FALLBACK_VERSION = '1.18.30';
-const OPENCODE_UA_VERSION_TTL_MS = 12 * 60 * 60 * 1000;
-let openCodeUaVersion = OPENCODE_UA_FALLBACK_VERSION;
-let openCodeUaVersionCheckedAt = 0;
-
-function refreshOpenCodeUaVersion(): void {
- if (process.env.VITEST) return;
- const now = Date.now();
- if (now - openCodeUaVersionCheckedAt < OPENCODE_UA_VERSION_TTL_MS) return;
- openCodeUaVersionCheckedAt = now; // stamp first so parallel calls share one lookup
- void fetch(OPENCODE_NPM_LATEST, { signal: AbortSignal.timeout(4000) })
- .then(res => (res.ok ? res.json() : null))
- .then((doc: unknown) => {
- const version = (doc as { version?: unknown } | null)?.version;
- if (typeof version === 'string' && /^\d+\.\d+\.\d+$/.test(version)) openCodeUaVersion = version;
- })
- .catch(() => { /* keep the last known version */ });
-}
-
-function openCodeZenHeaders(): Record<string, string> {
- refreshOpenCodeUaVersion();
- return {
- 'User-Agent': `opencode/${openCodeUaVersion} ai-sdk/provider-utils/4.0.23 runtime/bun/1.3.14`,
- 'x-opencode-client': 'cli',
- 'x-opencode-project': 'global',
- 'x-opencode-session': randomUUID(),
- 'x-opencode-request': randomUUID(),
- };
-}
+// Since 2026-09 the free roster is locked to the OpenCode client (#1249): every
+// free model answers 403 FreeTierError "OpenCode's free tier can only be used
+// from within OpenCode" on a valid key, and sending the OpenCode client headers
+// (#1204, reverted) does not change that. The catalog disables the free rows;
+// the provider stays registered for keys on OpenCode's paid models.
register(new OpenAICompatProvider({
platform: 'opencode',
name: 'OpenCode Zen',
baseUrl: 'https://opencode.ai/zen/v1',
- extraHeaders: openCodeZenHeaders,
}));
// OVHcloud AI Endpoints — OpenAI-compatible. Two free modes: anonymous
@@ -345,9 +308,12 @@ register(new OpenAICompatProvider({
// $0.0, please pay with fiat or send tao". The "free" tier requires a
// non-zero balance, which conflicts with the project's no-card criterion.
-// Reka — OpenAI-compatible (api.reka.ai/v1). Live-probed 2026-06-17: free via a
-// recurring monthly credit grant (no card; key from platform.reka.ai), billed
-// calls succeed with no 402. The OpenAI-compatible /v1/models lists two models:
+// Reka — OpenAI-compatible (api.reka.ai/v1). No longer free for new accounts
+// (#1202): Reka's FAQ now requires prepaid credits and a $0 balance answers
+// P001 Insufficient Balance. Accounts that still hold credit keep working
+// (both models answered 200 on 2026-09-17), so the provider stays registered;
+// the old `reka-flash` id 404s and is disabled in the catalog.
+// The OpenAI-compatible /v1/models lists two models:
// reka-flash-3 (text reasoning) and reka-edge-2603 (natively multimodal —
// accepts image/video input). Balance is dashboard-only (no credits API).
// Catalog rows live in the catalog (premium → age into free); they are NOT
diff --git a/server/src/providers/openai-compat.ts b/server/src/providers/openai-compat.ts
index 6c40372e..c5aff35d 100644
--- a/server/src/providers/openai-compat.ts
+++ b/server/src/providers/openai-compat.ts
@@ -68,7 +68,7 @@ export class OpenAICompatProvider extends BaseProvider {
readonly platform: Platform;
readonly name: string;
private readonly baseUrl: string;
- private readonly extraHeaders: Record<string, string> | (() => Record<string, string>);
+ private readonly extraHeaders: Record<string, string>;
private readonly validateUrl?: string;
/** Per-provider HTTP timeout override. OpenAI-compatible gateways often buffer
* non-streaming responses until generation completes, and reasoning models can
@@ -86,7 +86,7 @@ export class OpenAICompatProvider extends BaseProvider {
platform: Platform;
name: string;
baseUrl: string;
- extraHeaders?: Record<string, string> | (() => Record<string, string>);
+ extraHeaders?: Record<string, string>;
validateUrl?: string;
timeoutMs?: number;
keyless?: boolean;
@@ -169,14 +169,6 @@ export class OpenAICompatProvider extends BaseProvider {
return this.keyless ? {} : { 'Authorization': `Bearer ${apiKey}` };
}
- /** Extra headers for one upstream request. The function form is evaluated
- * once per request so a gateway that expects fresh per-request ids (OpenCode
- * Zen's `x-opencode-session`) does not have to live with values frozen at
- * registration time. */
- private requestHeaders(): Record<string, string> {
- return typeof this.extraHeaders === 'function' ? this.extraHeaders() : this.extraHeaders;
- }
-
/** Requesty's Leanstral route rejects greedy sampling when temperature=0.
* Omitting that value and supplying a neutral top_p keeps the caller's intent
* deterministic enough while using the provider's supported sampling path. */
@@ -286,7 +278,7 @@ export class OpenAICompatProvider extends BaseProvider {
headers: {
...this.authHeader(apiKey),
'Content-Type': 'application/json',
- ...this.requestHeaders(),
+ ...this.extraHeaders,
},
body: JSON.stringify({
model: modelId,
@@ -415,7 +407,7 @@ export class OpenAICompatProvider extends BaseProvider {
headers: {
...this.authHeader(apiKey),
'Content-Type': 'application/json',
- ...this.requestHeaders(),
+ ...this.extraHeaders,
},
body: JSON.stringify({
model: modelId,
@@ -524,7 +516,7 @@ export class OpenAICompatProvider extends BaseProvider {
method: 'GET',
headers: {
...this.authHeader(apiKey),
- ...this.requestHeaders(),
+ ...this.extraHeaders,
},
// 'request' bounds: a catalog body that hangs mid-transfer must not
// stall the health cycle past the deadline.
diff --git a/shared/types.ts b/shared/types.ts
index f390441b..3305ec4e 100644
--- a/shared/types.ts
+++ b/shared/types.ts
@@ -121,8 +121,8 @@ export type Platform =
// platform.agnes-ai.com (no card).
| 'agnes'
// Reka — OpenAI-compatible. Native multimodal models (reka-edge takes
- // image/video); free via a recurring monthly credit grant, key from
- // platform.reka.ai (no card).
+ // image/video). New accounts need prepaid credits (#1202); key from
+ // platform.reka.ai.
| 'reka'
// SiliconFlow — OpenAI-compatible. Registered for its FREE generative-media
// models (FLUX.1-schnell image, CosyVoice2 TTS) routed via services/media.ts;