# User statistics
- Active users: 244
- Expired users (tokens still in memory): 4
Service info: https://smolproxy.org
New beta proxy: https://beta.smolproxy.org
Smolproxy is a service that provides API access to multiple LLMs. There are a lot of uses for LLMs, a few examples:
You can use Smolproxy with any application/library/frontend where it is possible to specify a custom endpoint URL. Just to name a few:
Current stable providers: OpenAI, Gemini (incl. Vertex), Qwen, Moonshot (Kimi), Deepseek.
There are no refunds. Do not buy just for Claude access - it's rarely available.
For example, if the software that you use supports a custom OpenAI base URL and you want to use Smol, you can simply set the base URL to https://smolproxy.org/proxy/openai/v1 and set the API key to your token. See the full list of available API endpoints.
If you have any questions regarding service usage or payment, feel free to contact me.
Supports OpenAI's Responses API, supports Anthropic API for Deepseek/GLM/Kimi/Qwen.
Instructions for some of the popular harnesses:
Codex
Install Codex, don't launch it.
Create the folder ~/.codex, create ~/.codex/config.toml with:
model_provider = "smol"
[model_providers.smol]
name = "OpenAI"
base_url = "https://beta.smolproxy.org/openai/v1"
wire_api = "responses"
experimental_bearer_token = "your token here"
The IDE extension and the Codex app will also pick up the same config.
Claude Code
Install Claude Code, don't launch it.
Create ~/.claude.json with:
{"hasCompletedOnboarding": true}
~/.claude, create and edit ~/.claude/settings.json with:{
"env": {"ANTHROPIC_BASE_URL": "https://beta.smolproxy.org/anthropic"},
"apiKeyHelper": "echo YOURTOKENHERE"
}
Claude Code endpoints for other models:
https://beta.smolproxy.org/deepseek/anthropichttps://beta.smolproxy.org/glm/anthropichttps://beta.smolproxy.org/kimi/anthropicJune 20, 2026 - Some issues with Gemini on the proxy. Added 3 days to all tokens. Gemini API has issues currently, but if you're getting blank responses on /google/vertex, try disabling streaming to see the actual error, most likely PROHIBITED_CONTENT or similar.
June 16, 2026 - Added GLM 5.2.
May 23, 2026 - Added Alibaba with Qwen3.7 Max and some of their other hosted models. They also host Kimi K2.6, Deepseek V4 Pro, GLM 5.1.
May 22, 2026 - The main proxy will be switched to the software currently running on the beta proxy in the start of June.
Apr 24, 2026 - Added Kimi models to the beta proxy, contact me if you have any issues with it.
Apr 24, 2026 - Added Deepseek V4 Flash and Pro to the beta proxy.
Mar 17, 2026 - Added GPT 5.4 mini and nano. Added 4-week token renewal/purchase options.
Mar 5, 2026 - Added GPT 5.4. Use the new beta proxy for best results with the Responses API.
Feb 26, 2026 - Added Gemini 3.1 Flash Image (Nano Banana 2).
Feb 24, 2026 - Added GPT 5.3 Codex.
Feb 19, 2026 - Added Gemini 3.1 Pro.
Feb 12, 2026 - Added GLM 5 explicitly to both proxies (you could already use it on the beta proxy before).
Feb 6, 2026 - Launched the new proxy implementation at https://beta.smolproxy.org/, please test. Has some Claude.
Jan 12, 2026 - Old domain https://smol.services seems to have been disabled, switched to the new domain for now: https://smolproxy.org
Dec 28, 2025 - Proxy was down for ~2 hours due to issues with the new VPS host. Added 1 day to all tokens as compensation.
Dec 22, 2025 - Added GLM 4.7.
Dec 17, 2025 - Added Gemini 3 Flash.
Dec 12, 2025 - Added GPT-5.2.
Nov 21, 2025 - Added https://gen.smol.services - small frontend for Gemini 3 Pro image gen.
Nov 18, 2025 - Added Gemini 3 Pro (preview) to the proxy.
Nov 12, 2025 - Added GPT-5.1 to the proxy.
Oct 3, 2025 - Added GLM (ported from reanon, thanks for the implementation).
Sep 24, 2025 - Added Grok 4 Fast and Grok Code Fast 1.
/proxy/openai/proxy/google-ai/proxy/mistral-ai/proxy/deepseek, deepseek-reasoner is Deepseek V3.2 (thinking) and deepseek-chat is Deepseek V3.2. Prefills work automatically (last assistant message - Claude-style)./proxy/glm - thinking is disabled by default. To enable it in SillyTavern, add reasoning_effort: "high" in "Additional Parameters" -> "Include Body Parameters". For other frontends/scripts, you can either use the GLM-specific "thinking": {"type": "enabled"}, or pass the OpenAI-style reasoning_effort with low, medium or high./proxy/xaiSmall frontend for the Gemini 3 Pro image gen model (also called Nano Banana Pro)
You can use it with your smolproxy.org token or with a Gemini API key. Image handling and API requests are done directly from the browser, there's no backend.
Items might be out of stock because I want to limit the total amount of users.
After the order is done, you will automatically receive a newly-created temporary token that will work for the stated duration.
Crypto only. ETH (only through the main Ethereum chain), USDT, LTC, BTC, XMR. Due to small amounts, be careful about high fees, especially for BTC and sometimes ETH. Use LTC and XMR whether possible (longer confirmation but very low fees)
If you break the fair use limits (excessive usage 24/7, token sharing, etc), you will get banned.
Token renewal is completely automatic, click here to proceed
For expired tokens, the new expiration is set to current time plus renewal duration. For active tokens, renewal duration is added to the existing expiration date.
Examples:
You can renew the same token multiple times (e.g. buy the 2-week renewal two times), there's no upper limit. But only do this if truly needed, as there are no special discounts, guarantees or perks available for such cases.
Curent stable proxy: 30M tokens/day for Gemini, unlimited with fair use for others.
New beta proxy - $100 API credits/day with accurate cost tracking, including caches. Specific quotas:
Can I renew an expired token?
Yes, but only if it's still in the memory. All tokens that have expired will be purged after roughly 1 week of inactivity. You can check if your token still works by going to https://smolproxy.org/user/lookup.
How long does it take you to restore service if some LLM endpoint stops working?
If an LLM endpoint stops working and I'm able to restore it - I will. If it takes me too long, I will compensate for the inconvenience by extending the duration of all tokens (including those that expired during the outage) by at least the outage duration (usually more). For example, if the Deepseek endpoint didn't work (not due to upstream issues) for for 6 hours and I was able to restore it, I will extend all tokens by at least 6 hours.
4chan tripcode: !!ahg+yVFXVZL (previous tripcode before the 4chan incident: !!zLqNx4NnbXf)