GLM 5.3 and GLM 5.3 Flash Uncensored are live

Two new models are available on XField today: GLM 5.3 Uncensored and GLM 5.3 Flash Uncensored. Both are uncensored versions of Z.ai's GLM 5.3 family, with refusal behavior removed and no moderation layer added on our side. Both take 1M tokens of context. They split the work cleanly: the full model is Z.ai's coding and cybersecurity model with a 128K output limit, and Flash is the lighter one for volume.

If you already have a key, there is nothing to set up. Change the model id and send the request.

The two models side by side

Both models work on all three protocols: OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages. Pricing is shared with your invitation.

What GLM 5.3 is for

GLM 5.3 is the model to reach for when the answer is long. Its 128K output limit, the longest we serve, means it can write the whole thing rather than a summary of the thing. Z.ai tuned it for coding and cybersecurity, and on most benchmarks it sits just behind DeepSeek V4.1 Flash, which makes it a strong second opinion from a different model family.

• Security research. Walk through a vulnerability class, reason about exploitability in your own lab, draft detection logic and the test cases that prove it fires.
• Red-team planning. Lay out an engagement against your own systems: scope, objectives, techniques to exercise, what each phase should prove.
• Long reports. Feed it a full codebase, a stack of logs, or an incident timeline and get back a structured write-up in one response.

It answers the question you asked. It does not stop halfway to explain why it would rather not.

What GLM 5.3 Flash is for

Flash is the lighter GLM. It is built for jobs where you send a lot of text, need a reasonable answer for each item, and care about the total bill more than the last few points of quality.

• Datasets. Generate synthetic examples, including the adversarial and unpleasant ones a filtered model refuses to write.
• Labeling. Classify content for trust and safety work without the model refusing to look at the content.
• Batch jobs. Summaries, extraction, and reformatting across thousands of documents.

For large batches, start with low reasoning effort. Raise it only for the items that come back wrong. Most routine work does not need the model to think hard, and you pay for the tokens it spends thinking.

Which one to pick

Pick GLM 5.3 when:

• The output is long. Anything past 64K tokens needs the full GLM 5.3.
• You want a second opinion on work DeepSeek V4.1 Flash already did.
• You run a few important requests rather than many routine ones.

Pick GLM 5.3 Flash when:

• You run the same prompt shape over many inputs.
• Input volume dominates your token count.
• Good enough per item is fine, and you can rerun the failures.

For coding agents, check DeepSeek V4.1 Flash first. Agents resend the same large context on every turn, so most of their input is cached, and deepseek-v4.1-flash-1017-uncensored is usually the cheaper choice there. See the models page for the full comparison.

Calling it from the Messages API

Both models work with the Anthropic SDK. Point it at https://xfield.cloud.brutal.ai/api, pass your key, and turn thinking on. You get Anthropic thinking blocks back, followed by the answer.

import os
import anthropic
 
client = anthropic.Anthropic(
    base_url="https://xfield.cloud.brutal.ai/api",
    api_key=os.environ["XFIELD_API_KEY"],
)
 
message = client.messages.create(
    model="glm-5.3-uncensored",
    max_tokens=16000,
    thinking={"type": "enabled", "budget_tokens": 8000},
    messages=[
        {
            "role": "user",
            "content": "Write Sigma rules that detect credential dumping from LSASS "
                       "in my lab, with a test case for each rule.",
        }
    ],
)
 
for block in message.content:
    if block.type == "thinking":
        print("[thinking]", block.thinking)
    elif block.type == "text":
        print(block.text)

import Anthropic from "@anthropic-ai/sdk";
 
const client = new Anthropic({
  baseURL: "https://xfield.cloud.brutal.ai/api",
  apiKey: process.env.XFIELD_API_KEY,
});
 
const message = await client.messages.create({
  model: "glm-5.3-uncensored",
  max_tokens: 16000,
  thinking: { type: "enabled", budget_tokens: 8000 },
  messages: [
    {
      role: "user",
      content:
        "Write Sigma rules that detect credential dumping from LSASS " +
        "in my lab, with a test case for each rule.",
    },
  ],
});
 
for (const block of message.content) {
  if (block.type === "thinking") console.log("[thinking]", block.thinking);
  else if (block.type === "text") console.log(block.text);
}

To use Flash, swap the model id for glm-5.3-flash-uncensored. To set reasoning effort on the Messages API, pass output_config with an effort such as "low" or "high". The budget_tokens value is accepted for compatibility but is not enforced as a separate budget.

Get access

XField is invite-only. Request access, and once you are in, create a key under Settings › API keys. Compare every model, including both GLMs, on the models page.