One endpoint.
Three protocols.

SuhadaCosme serves China's frontier open-weight models through a single API that speaks OpenAI, Anthropic and Gemini natively. Point any existing client, SDK or agent harness at a new base URL — that's the whole migration. One key works for every model and every protocol.

01Get your key

Keys are provisioned personally during beta. Email hello@suhadacosme.com with a sentence about your workload — you're usually running within a day.

02Make your first call

curl https://api.suhadacosme.com/v1/chat/completions \
  -H "Authorization: Bearer $SUHADA_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3",
    "messages": [
      {"role": "user", "content": "Say hello in five languages."}
    ]
  }'

03Point your stack at it

Authentication

One key, three native header styles. Use whichever your client already sends — no adapter, no shim.

Models

The same model ID string works across all three protocols. Prices are USD per 1M tokens, cache-miss input / output.

List models over the API: GET /v1/models with your Bearer key.

OpenAI API protocol

Full Chat Completions compatibility. Any OpenAI SDK, in any language, works by changing base_url.

·Endpoints

·curl

curl https://api.suhadacosme.com/v1/chat/completions \
  -H "Authorization: Bearer $SUHADA_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-pro",
    "messages": [{"role": "user", "content": "hello"}],
    "stream": false
  }'

·Python SDK

from openai import OpenAI

client = OpenAI(
    base_url="https://api.suhadacosme.com/v1",
    api_key="sk-...",
)

resp = client.chat.completions.create(
    model="glm-5.3",
    messages=[{"role": "user", "content": "hello"}],
)
print(resp.choices[0].message.content)

·JavaScript / TypeScript SDK

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.suhadacosme.com/v1",
  apiKey: process.env.SUHADA_KEY,
});

const resp = await client.chat.completions.create({
  model: "kimi-k3",
  messages: [{ role: "user", content: "hello" }],
});
console.log(resp.choices[0].message.content);

Anthropic API protocol

Native Messages API — drop-in for Claude SDKs and Claude Code itself. Send x-api-key plus anthropic-version.

·Endpoints

·curl

curl https://api.suhadacosme.com/v1/messages \
  -H "x-api-key: $SUHADA_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.2",
    "max_tokens": 1024,
    "system": "You are a precise senior engineer.",
    "messages": [{"role": "user", "content": "hello"}]
  }'

·Python SDK

from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.suhadacosme.com",
    api_key="sk-...",
)

msg = client.messages.create(
    model="glm-5.3",
    max_tokens=1024,
    messages=[{"role": "user", "content": "hello"}],
)
print(msg.content[0].text)

Gemini API protocol

Native generateContent surface for Gemini / Vertex-shaped clients — no rewriting your pipeline.

·Endpoints

·curl

curl "https://api.suhadacosme.com/v1beta/models/kimi-k3:generateContent" \
  -H "x-goog-api-key: $SUHADA_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{"parts": [{"text": "hello"}]}]
  }'

·JavaScript SDK (@google/genai)

import { GoogleGenAI } from "@google/genai";

const ai = new GoogleGenAI({
  apiKey: process.env.SUHADA_KEY,
  httpOptions: { baseUrl: "https://api.suhadacosme.com" },
});

const resp = await ai.models.generateContent({
  model: "glm-5.3",
  contents: "hello",
});
console.log(resp.text);

The ?key= query parameter is accepted as an alternative to the x-goog-api-key header, matching Google's client conventions.

Streaming

All three protocols stream with server-sent events, in each protocol's native chunk shape.

• OpenAI — "stream": true; delta chunks terminate with data: [DONE]
• Anthropic — "stream": true; events from message_start through content_block_delta to message_stop
• Gemini — :streamGenerateContent?alt=sse; JSON chunks in data: frames

Errors

Errors follow each protocol's native schema, so your existing error handling keeps working.

{
  "error": {
    "message": "Model glm-9 does not exist",
    "type": "invalid_request_error",
    "code": "model_not_found"
  }
}

Default rate limits are generous and scale with your plan. If you're hitting 429s, tell us your target RPM and we'll size your key accordingly.

Plug into your harness.

Agent harnesses and coding tools already speak one of our three protocols. Each guide below is a complete, verified configuration.

Claude Code

Claude Code talks the native Anthropic protocol — point it at SuhadaCosme with two environment variables.

·Environment variables

export ANTHROPIC_BASE_URL="https://api.suhadacosme.com"
export ANTHROPIC_AUTH_TOKEN="sk-..."

# optional — pin the model (default follows your Claude Code settings)
export ANTHROPIC_MODEL="glm-5.3"

·Or persist it in settings

{
  "env": {
    "ANTHROPIC_BASE_URL": "https://api.suhadacosme.com",
    "ANTHROPIC_AUTH_TOKEN": "sk-...",
    "ANTHROPIC_MODEL": "glm-5.3"
  }
}

·Run

claude
# inside the session, switch models any time:
#   /model kimi-k3        (SWE specialist)
#   /model deepseek-v4-pro (deep reasoning, thin bill)

Codex CLI

Codex supports any OpenAI Chat-Completions-compatible provider via model_providers.

·~/.codex/config.toml

# ~/.codex/config.toml
model = "suhada/glm-5.3"
model_provider = "suhada"

[model_providers.suhada]
name = "SuhadaCosme"
base_url = "https://api.suhadacosme.com/v1"
env_key = "SUHADA_API_KEY"
wire_api = "chat"

·Run

export SUHADA_API_KEY="sk-..."
codex                      # uses suhada/glm-5.3 from config
codex --model suhada/kimi-k3  # override per-run

OpenClaw

Add SuhadaCosme as a custom provider in openclaw.json (legacy path ~/.clawdbot/clawdbot.json is symlinked automatically).

·~/.openclaw/openclaw.json

{
  "models": {
    "providers": {
      "suhada": {
        "baseUrl": "https://api.suhadacosme.com/v1",
        "apiKey": "sk-...",
        "api": "openai-completions",
        "models": [
          { "id": "glm-5.3" },
          { "id": "glm-5.2" },
          { "id": "kimi-k3" },
          { "id": "deepseek-v4-pro" }
        ]
      }
    }
  }
}

·Select the model

openclaw models set suhada/glm-5.3
# or pick it interactively from the models list

Cline / Roo Code

Both VS Code extensions have a built-in "OpenAI Compatible" provider — no config files needed.

• API Provider — choose OpenAI Compatible in the extension's settings panel
• Base URL — https://api.suhadacosme.com/v1
• API Key — your SuhadaCosme key
• Model ID — e.g. glm-5.3 or kimi-k3

Roo Code follows the same flow in its provider dropdown. Enable streaming in the panel for long generations.

aider

aider's OpenAI-compatible mode takes the base URL as a flag or environment variable.

export OPENAI_API_BASE="https://api.suhadacosme.com/v1"
export OPENAI_API_KEY="sk-..."

aider --model openai/glm-5.3
# strong SWE pair:
aider --model openai/kimi-k3 --edit-format diff

Hermes

Hermes speaks the OpenAI protocol for model backends. In your provider/model config, set:

• Endpoint / Base URL — https://api.suhadacosme.com/v1
• API key — your SuhadaCosme key, passed as Bearer
• Model — any catalog ID, e.g. glm-5.2 for long-horizon agent loops

Anything else that speaks OpenAI, Anthropic or Gemini — LangChain, LlamaIndex, Continue, your own backend — works the same way: base URL + key + model ID.

FAQ

• One key for all protocols? — Yes. The same key authenticates OpenAI, Anthropic and Gemini surfaces.
• Same model across protocols? — Yes, identical model IDs and identical serving paths; pick the protocol your client already speaks.
• Tool calling / function calling? — Supported through each protocol's native tool-use shapes.
• Which model should I start with? — glm-5.3 for agentic coding, deepseek-v4-pro for deep reasoning at minimum cost, deepseek-v4-flash for volume.
• Need a limit raised, a second key, or a model not listed? — hello@suhadacosme.com — we're fast.