1. 7API
  2. Models
  3. GLM
  4. GLM 5.3 Abliterated

GLM 5.3 Abliterated API

Model ID: glm-5.3-abliterated · Developer: Zhipu AI · Available on 7API: yes, as of 2026-11-20

The flagship abliterated model on 7API, with refusal behavior removed at the weight level. Thinking is always on, at high effort or above, and the default output budget is 130,000 tokens to leave room for long reasoning.

GLM 5.3 Abliterated is an abliterated version of GLM 5.3, available through 7API's uncensored AI API. Call GLM 5.3 Abliterated with any OpenAI- or Anthropic-compatible client. Use base URL https://7api.st/v1 (OpenAI) or https://7api.st (Anthropic) and send "model": "glm-5.3-abliterated".

Specifications

Model IDglm-5.3-abliterated
Context window1,000,000 tokens
Max output130,000 tokens
InputText
Tool callingYes
StreamingYes
Reasoning efforthigh, xhigh, max (default high)
API formatsChat Completions, Responses, Anthropic Messages

Pricing

USD per 1M tokens
Input$0.80
Output$0.80
Cached input$0.25

For example, an agent turn with 150,000 cached input tokens, 2,000 new input tokens and 800 output tokens costs about $0.0397 on GLM 5.3 Abliterated. Billing is pay per token from a prepaid balance. See Pricing and billing.

Benchmarks

Independent leaderboards don't test abliterated builds. The base model, GLM 5.3, scores 45 on the Artificial Analysis Intelligence Index at max effort. Expect GLM 5.3 Abliterated to be close to that on general tasks, and test it on your own prompts.

GLM 5.3 Abliterated vs. other GLM models

ModelTypeBased onInput / 1MOutput / 1MCached / 1MMax outputReasoning effort
GLM 5.3Standard—$0.05$0.05$0.0166Klow high xhigh max
GLM 5.3 UncensoredUncensoredGLM 5.3$0.05$0.05$0.0166Klow high xhigh max
GLM 5.3 Flash UncensoredUncensoredGLM 5.3 Flash$0.025$0.025$0.002566Klow high xhigh max
GLM 5.3 AbliteratedAbliteratedGLM 5.3$0.80$0.80$0.25130Khigh xhigh max

Example

Python (OpenAI SDK)
from openai import OpenAI

client = OpenAI(base_url="https://7api.st/v1", api_key="YOUR_7API_KEY")
r = client.chat.completions.create(
    model="glm-5.3-abliterated",
    messages=[{"role": "user", "content": "Write a haiku about caching."}],
)
print(r.choices[0].message.content)

In Claude Code: /model claude-glm-5.3-abliterated. Setup guides: Claude Code, OpenCode, Codex, Cursor.

What is GLM 5.3 Abliterated?

GLM 5.3 Abliterated is Zhipu AI's GLM-5.3 with its refusal direction removed from the weights (abliteration). It keeps GLM-5.3's reasoning, tool calling and 1M-token context, but declines far fewer requests. See what abliteration is.

How do I use GLM 5.3 Abliterated through an API?

Send "model": "glm-5.3-abliterated" to https://7api.st/v1/chat/completions with your 7API key, using any OpenAI-compatible SDK. Anthropic-format clients use https://7api.st, and Claude Code uses /model claude-glm-5.3-abliterated.

GLM 5.3 Abliterated vs GLM 5.3 Uncensored: which should I use?

GLM 5.3 Abliterated refuses least and has a 130,000-token output budget. GLM 5.3 Uncensored costs much less and is better for high-volume agent work. See the comparison table above.

Can I run GLM 5.3 Abliterated in Claude Code or OpenCode?

Yes. In Claude Code run /model claude-glm-5.3-abliterated. The OpenCode config already includes it with high, xhigh and max effort variants.

Does GLM 5.3 Abliterated support reasoning?

Yes. Thinking is always on. Effort defaults to high, and you can raise it to xhigh or max with reasoning_effort. Lower values are raised to high.

Also searched as: GLM 5.3 Abliterated, GLM-5.3 Abliterated, glm-5.3-abliterated, abliterated GLM 5.3.