GLM-5.3 Flash API: Pricing and a Free Start

Prices per 1M tokens as of November 2, 2026:

On OpenRouter, GLM-5.3 Flash has more than thirty providers, and their prices differ many times over: many stick to Z.ai's price of $0.15 and $0.50, while the cheapest one, Relace, charges $0.0285 and $0.0923. Compare with it: the difference against the cheapest provider of the same model is about ×5, and against Z.ai's API about ×24 on input and ×27 on output.

At Z.ai and many providers, the resent part of a request — conversation history, system prompt — is read from cache cheaper than regular input. JoinGonka has no separate cache price; all input costs the same. But cached input at Z.ai still costs $0.03 per million — many times more than our regular input.

The JoinGonka price on this page is live: it comes from the gateway API and changes with the GNK rate. What makes it low is how the Gonka network works: the price is set there by the settlement for the compute work of hosts, not by the model developer's price list.

1. Key. Sign up at JoinGonka Gateway and create a jg-… key in your dashboard. New accounts receive 3M free tokens.

2. Request. The OpenAI-compatible API address is https://gate.joingonka.ai/v1, the model ID is zai-org/GLM-5.3-Flash:

curl https://gate.joingonka.ai/v1/chat/completions \
  -H "Authorization: Bearer jg-your-key" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "zai-org/GLM-5.3-Flash",
    "max_tokens": 2000,
    "messages": [{"role": "user", "content": "Find the bug: def avg(xs): return sum(xs) / len(xs) - 1"}]
  }'

The same in Python with the openai library:

from openai import OpenAI

client = OpenAI(api_key="jg-your-key", base_url="https://gate.joingonka.ai/v1")
response = client.chat.completions.create(
    model="zai-org/GLM-5.3-Flash",
    max_tokens=2000,
    messages=[{"role": "user", "content": "Explain in two sentences what a race condition is"}],
)
print(response.choices[0].message.content)

3. Coding tools. The installer will write the gateway and model into the config of Claude Code, Codex CLI, OpenCode and other tools with a single command — it will ask you for the tool and the key itself:

npx @joingonka/setup --model glm

For Anthropic Messages API clients the address is https://gate.joingonka.ai, without /v1. Step-by-step guides for each tool are in the Tools section.

• Free tokens. A new account gets 3M tokens — that's hundreds of requests to GLM-5.3 Flash: enough to test the model on your own tasks and tune your reasoning budget. No card or crypto needed for that.
• Chat without registration. In the free chat on our site you can pick GLM-5.3 Flash and watch it reason: the chain of thought is shown in a separate collapsible block.
• Then — pay per token. There's no subscription: you top up your balance in GNK, USDT or by card, actual tokens are deducted, and daily usage is visible in your dashboard.

Z.ai gives away its older lightweight models for free — GLM-4.7-Flash and GLM-4.5-Flash. GLM-5.3 Flash is paid there, so the way to try that exact model without paying is with JoinGonka's starter tokens.

GLM-5.3 Flash is the lightweight model in the GLM-5.3 lineup from Z.ai, the international brand of Beijing-based Zhipu AI. Lightweight is relative: it's a 320-billion-parameter MoE model, of which roughly 18 billion are active per token. Attention is hybrid — some layers are sparse, some are linear — so long context costs it less. The weights are open under the MIT license. The working parameters in JoinGonka are set by the deployment on the network's hosts:

• Context — 390,000 tokens. That's the verified-by-queries minimum: Z.ai and OpenRouter providers offer a window of up to 1M tokens, the network's hosts are configured for 400,000, and we publish only what's confirmed.
• Response — up to 8,192 tokens, including reasoning.
• Reasoning arrives in a separate reasoning_content field, tool calls come through the standard tool_calls, including the second and subsequent rounds; JSON mode (response_format) works.
• Text only: the network doesn't accept images yet.

The models, their IDs, and limits are in the Gonka API documentation, availability and latency are on the gateway status page.

GLM-5.3 Flash reasons first, then answers, and both stages count toward a single output limit. Even for a simple question, reasoning takes 250–450 tokens; for a substantive task, thousands. So the familiar max_tokens: 256 yields finish_reason: "length" and an empty content: the model spent the limit on reasoning and never got to the answer.

Reasoning is billed as output tokens. If you don't need it at all, another model on the network is handier for short, fast answers. In agents like Cline and Cursor, check the output limit for auxiliary requests — context compression, conversation titles: with a limit of a couple hundred tokens, GLM-5.3 Flash will return an empty response.

Both models are available in JoinGonka at the same price, so choose by task, not by budget:

In short: GLM-5.3 Flash is for when you need to think through complex logic and don't need a long answer; DeepSeek V4 Flash is for long answers, large documents, and agents that run for hours. The network's third model, MiniMax M2.7, suits fast short answers, but MiniMax itself is retiring it from its API. Switching between models is one model line in the request.