tokenclub - Open-source LLMs, served at the cheapest rates

The model catalog

Frontier open models, one base URL

Every model is deployed, quantized and load-balanced by us - flagship quality at the best price per token on the market, and 90% off every cached token. No GPU babysitting. Text only.

DeepSeek V4 Flash

284B · MoE

DeepSeek's speed-tuned model for high-throughput coding and chat at rock-bottom cost across a 1M-token window.

$0.08 / $0.17

in · out per M

cached in $0.008 · 90% off

32 t/sthroughput

DeepSeek V4 Pro

1.6T · MoE

DeepSeek's flagship for deep reasoning, full-codebase analysis, and long-horizon agents.

$0.425 / $0.86

in · out per M

cached in $0.043 · 90% off

54 t/sthroughput

GLM 5.2

355B · MoE

Flagship reasoning and agentic model, tuned for tool-calling and long-horizon coding tasks.

$0.75 / $2.41

in · out per M

cached in $0.075 · 90% off

57 t/sthroughput

MiniMax M3

428B · MoE

Sparse-attention model for cheap, fast long-context work at scale.

$0.23 / $0.95

in · out per M

cached in $0.023 · 90% off

54 t/sthroughput

Kimi K3

2.8T · MoE

Moonshot's open-weight reasoner, strong on large-repo navigation, debugging, and multi-step tool use.

$2.00 / $10.00

in · out per M

cached in $0.50 · 75% off

35 t/sthroughput

gpt-oss-120b

117B · MoE

OpenAI's open-weight model with configurable reasoning depth and native tool use, runs lean and fast.

$0.02 / $0.16

in · out per M

cached in $0.002 · 90% off

37 t/sthroughput

Drop-in replacement

Change two lines. Keep your whole stack.

tokenclub speaks the OpenAI API you already use. Swap the base URL and key - every SDK, framework and tool just works. No rewrites, no vendor lock-in.

Same schema - chat, completions, embeddings, streaming, function calling.
Works everywhere - LangChain, LlamaIndex, Vercel AI SDK, the raw HTTP call.
Super fast TTFT - global GPU regions with automatic failover.

curl https://api.tokentoken.club/v1/chat/completions \
  -H "Authorization: Bearer $TOKENCLUB_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "moonshotai/kimi-k3",
    "messages": [
      {"role": "user", "content": "Explain MoE in one line."}
    ],
    "stream": true
  }'

from openai import OpenAI

client = OpenAI(
    base_url="https://api.tokentoken.club/v1",
    api_key=os.environ["TOKENCLUB_KEY"],
)

stream = client.chat.completions.create(
    model="deepseek-v3",
    messages=[{"role": "user", "content": "Explain MoE."}],
    stream=True,
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="")

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.tokentoken.club/v1",
  apiKey: process.env.TOKENCLUB_KEY,
});

const res = await client.chat.completions.create({
  model: "qwen-3",
  messages: [{ role: "user", content: "Explain MoE." }],
});

console.log(res.choices[0].message.content);

Open-source AI,
served at the
best rates

Frontier open models, one base URL

Change two lines. Keep your whole stack.

Welcome to the Token Club