Welcome to A2Agent: The Cheapest AI Model API for Open-Source Models

A2Agent Team · 2026-10-17T00:00:00Z

A2Agent is the cheapest AI model API for the top open-source model families: DeepSeek, Qwen, GLM, Kimi and MiniMax. You sign up once, create one API key, and call any of 22 models through the OpenAI, Anthropic or Gemini API formats you already use. Models cost up to 50% less than the vendor's own list price, and you pay only for the tokens you use.

Open-source models have caught up fast. DeepSeek, Qwen, GLM, Kimi and MiniMax now handle coding, agents, long documents and vision work that used to need a closed frontier model, at a fraction of the price. A2Agent puts all of them behind one account and makes them cheaper still.

A2Agent is run by Omnimodel Technology Limited, a Hong Kong company.

At a glance

  • 22 models from 5 open-source model families, all behind one key
  • OpenAI, Anthropic and Gemini compatible APIs, so existing SDKs and tools work with a base URL change
  • Prices up to 50% below official rates
  • $5 free credit at signup, no card required
  • Pay as you go from $1, with no top-up fee, no monthly fee and no expiry while your account is active
  • Works with Claude Code, Codex CLI, Cline, Roo Code, Cursor, Zed and more than 30 other coding tools

Open-source models on A2Agent

Every model on A2Agent comes from the vendor's official platform. When an upstream account has trouble, we route the request to another of our accounts with the same vendor, so you keep getting the model you asked for.

Vendor Models
DeepSeek V4 Flash, V4.1 Flash, V4 Pro
Qwen Qwen3.6 Flash, Qwen3.7 Flash, Qwen3.8 Flash, Qwen3.5 Plus, Qwen3.7 Plus, Qwen3.7 MAX, Qwen3.8 MAX
GLM (Z.ai) GLM-5, GLM-5.1, GLM-5.2, GLM-5.3, GLM-5.3 Flash
Kimi (Moonshot) Kimi K2.5, Kimi K2.6, Kimi K2.7 Code, Kimi K3
MiniMax MiniMax M2.5, M2.7, M3

Most have a 1M-token context window. The Flash models cost the least and handle everyday work; for harder reasoning, coding or vision tasks, look at DeepSeek V4 Pro, GLM-5.3, Qwen3.8 MAX or Kimi K3. The Models page lists each one with its context length, tags and price.

Pricing: the cheapest AI model API for open-source models

You pay per token, with input and output metered separately at each model's rate. Here are the ten cheapest models today, in US dollars per million tokens:

Model Official price (in / out) A2Agent price (in / out) Discount
DeepSeek V4 Flash $0.140 / $0.280 $0.084 / $0.168 40%
GLM-5.3 Flash $0.150 / $0.500 $0.105 / $0.350 30%
DeepSeek V4.1 Flash $0.170 / $0.630 $0.102 / $0.378 40%
Qwen3.8 Flash $0.150 / $0.470 $0.135 / $0.423 10%
Qwen3.7 Flash $0.200 / $0.800 $0.140 / $0.560 30%
MiniMax M2.5 $0.300 / $1.20 $0.150 / $0.600 50%
MiniMax M2.7 $0.300 / $1.20 $0.150 / $0.600 50%
MiniMax M3 $0.300 / $1.20 $0.150 / $0.600 50%
Qwen3.6 Flash $0.200 / $1.20 $0.140 / $0.840 30%
DeepSeek V4 Pro $0.435 / $0.870 $0.261 / $0.522 40%

The rate you pay depends on your API key's group, and the Pricing page always shows current prices. For a longer comparison aimed at coding work, see Cheapest AI Models for Code in 2026.

Credits and top-ups

New accounts get $5 in free credit, with no card required, usable on any model your key can reach. After that you top up a prepaid balance, starting at $1. There's no top-up fee, no monthly fee and no minimum spend, and your balance doesn't expire while your account is active.

Top-ups currently earn bonus credit:

Top-up First top-up Every later top-up
$10 +$10 none
$60 +$20 +$20
$100 +$30 +$30
$200 or more +$50 +$50

Unused purchased credit can be refunded within 7 days of purchase. Free credit and bonus credit can't be refunded. Teams that need volume discounts, dedicated concurrency or higher rate limits can contact us about an enterprise plan.

Connect your code

A2Agent speaks the OpenAI, Anthropic and Gemini API formats. If your code already uses the OpenAI SDK, change two settings, the base URL and the key:

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_API_KEY",
    base_url="https://api.a2agent.me/v1",
)

response = client.chat.completions.create(
    model="deepseek-v4-pro",
    messages=[{"role": "user", "content": "Hello"}],
)

print(response.choices[0].message.content)

To switch models, change the model name. Anthropic-format requests go to https://api.a2agent.me/v1/messages. The OpenAI SDK integration guide and the API docs cover the details.

Use it in your coding tools

Coding agents read and write a lot of tokens, so they're where lower prices save the most. For Claude Code, add this to ~/.claude/settings.json:

{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "YOUR_API_KEY",
    "ANTHROPIC_BASE_URL": "https://api.a2agent.me",
    "ANTHROPIC_MODEL": "deepseek-v4-pro",
    "CLAUDE_CODE_MAX_OUTPUT_TOKENS": "32000",
    "CLAUDE_CODE_DISABLE_UNKNOWN_MODEL_WINDOW_ENFORCEMENT": "1"
  }
}

Restart Claude Code and it runs on DeepSeek V4 Pro. For quick edits and debugging, deepseek-v4-flash is cheaper; for agent workflows with lots of tool calls, try minimax-m3.

Codex CLI connects through the OpenAI Responses API. Cline, Roo Code, Kilo Code, Continue, Aider, OpenCode, Zed, Goose, Crush, Factory Droid and others connect through OpenAI Chat Completions. Our Integrations page covers 35 tools and marks which ones accept a custom endpoint fully and which only partly.

Control spending per key

Each API key can carry its own limits, which is useful when you hand keys to teammates, scripts or agents:

  • a total quota in US dollars
  • spending caps per 5 hours, per day and per 7 days
  • an optional expiry date
  • an optional IP allowlist

Usage, charges and remaining balance are all visible in the dashboard. When a balance runs out, requests return an error until you top up, so nothing bills past what you've paid.

Get started

  1. Create an account. Your $5 credit is added automatically.
  2. Open API Keys in the console, click "Create Key", name it and pick its groups (selecting all groups is the simplest choice).
  3. Point your app or coding tool at https://api.a2agent.me and choose a model from the Models page.

If you get stuck, email support@a2agent.me or ask in our Discord or Telegram community, where we post model updates, example projects and integration help. You can also follow us on X and GitHub, and check service health on our public status page.