Z.ai
Operational· checked every few minutes

GLM-4.7 Flash Heretic API

The GLM-4.7 Flash Heretic API: a community uncensored GLM-4.7 Flash with its refusals removed, with reasoning, tools and a 200K-token context, 13% below list price

Price per 1M tokens13%off
Input
$0.061$0.07
Output
$0.35$0.40
Cached input
$0.0305$0.035
Get your API keyTry it in the playground
Model IDglm-4.7-flash-hereticUpdated
chat.py
from openai import OpenAI
client = OpenAI(
    base_url="https://api.zurelay.com/v1",
    api_key="YOUR_ZURELAY_KEY",
)
stream = client.chat.completions.create(
    model="glm-4.7-flash-heretic",
    messages=[
        {"role": "user", "content": "Explain quicksort briefly."},
    ],
    stream=True,
)
for chunk in stream:
    if chunk.choices:
        print(chunk.choices[0].delta.content or "", end="")

Base URL https://api.zurelay.com/v1. Works with any OpenAI SDK. Chat API docs

Context window
200K tokens
Max output
24K tokens
Reasoning
none · low · medium · high
Input
Text
Status
Operational

Pricing

GLM-4.7 Flash Heretic API pricing. A fraction of the list price.

Pay as you go from prepaid credit, with no subscription. Top up from $10. Requests that fail are never billed.

RateZurelayZ.ai list
Input
per 1M tokens
$0.061$0.07
Output
per 1M tokens
$0.35$0.40
Cached input
per 1M tokens
$0.0305$0.035

One rate at every prompt length, up to the full 200K context window. Streaming, tool calls and structured outputs cost nothing extra.

Every Uncensored LLM model and price
Savings calculator

Move the sliders to your monthly usage.

Input tokens per month50M
Output tokens per month10M
Zurelay
$6.55
Z.ai list price
$7.50
You save every month$0.95

$11.40 a year

Cost

GLM-4.7 Flash Heretic API cost, worked out.

Typical workloads at Zurelay's live prices, next to the list price.

  • Chatbot

    50,000 requests a month, 1,500 tokens in and 300 out, 30% of the input cached. $0.000183 a request.

    $9.14a month$10.46 at list

  • Documents

    20,000 requests a month, 8,000 tokens in and 500 out, 20% of the input cached. $0.000614 a request.

    $12.28a month$14.08 at list

  • Coding agent

    5,000 requests a month, 40,000 tokens in and 1,500 out, 85% of the input cached. $0.00193 a request.

    $9.64a month$11.05 at list

  • Classification

    1,000,000 requests a month, 400 tokens in and 5 out. $0.000026 a request.

    $26.15a month$30.00 at list

Overview

What is GLM-4.7 Flash Heretic?

GLM-4.7 Flash Heretic is a community uncensored version of Z.ai's GLM-4.7 Flash, made with Heretic, an open-source tool that removes a model's refusals automatically. It isn't made or endorsed by Z.ai. The Zurelay GLM-4.7 Flash Heretic API serves it through one OpenAI-compatible endpoint with a 200K-token context window, reasoning you can switch off and tool calling, at 13% below list price. Zurelay's Terms of Service still apply to everything you send it.

GLM-4.7 Flash Heretic starts from GLM-4.7 Flash, Z.ai's fast open-weight model, and removes its refusals with Heretic, an open-source tool that finds the direction in the model's internal activations linked to refusing and edits the weights to suppress it, without retraining. The result, published on Hugging Face by Olafangensan, answers prompts the original declines. It is a community build, not made or endorsed by Z.ai. It reads and writes text, with a context window of 200,000 tokens and up to 24,000 output tokens. On Zurelay the model ID is glm-4.7-flash-heretic, and olafangensan-glm-4.7-flash-heretic works too. It costs $0.061 per 1M input tokens and $0.35 per 1M output tokens, with cached input at $0.0305.

It suits work that filtered models often refuse or water down: fiction with dark themes, roleplay, red-team prompts for testing your own systems, and plain answers on sensitive topics. Removing refusals can shift other abilities too, so test it on your own tasks before you rely on it. Fewer refusals don't change what you may use it for. Zurelay's Terms of Service (https://zurelay.com/legal/terms) still apply to everything sent to it, including the acceptable use rules in section 7, and you are responsible for how you use its output.

It reasons before it answers, at low by default. Set reasoning_effort to none, low, medium or high: none switches reasoning off for a fast direct answer, and in our tests high thought more than twice as long as low on a hard question. It calls tools on its own, several at once and while streaming, follows json_object and json_schema in response_format, and takes stop sequences and the usual sampling settings; logprobs isn't supported. In our tests it found one fact buried in a 168K-token prompt in 44 seconds. You pay $0.061 per 1M input tokens and $0.35 per 1M output tokens against a list price of $0.07 and $0.40.

Strengths

Where GLM-4.7 Flash Heretic shines. And what teams build with it.

01

Answers instead of refusing

Heretic removes the learned refusal behavior, so fiction, red-team prompts and sensitive questions get an answer rather than a refusal.

02

Reasoning you control

It thinks at low by default; raise it to medium or high for hard problems, or set none for a fast direct answer.

03

Tools on its own

It decides when to call a tool, calls several at once, also while streaming, and reads the results, so it works in agents.

04

A low price per token

It costs $0.061 per 1M input tokens and $0.35 per 1M output tokens, and repeated prompts bill at the cached rate of $0.0305.

Use cases

  • Uncensored agents

    Agents that call tools and reason over several steps without refusing parts of the task.

  • Testing your own AI systems

    Adversarial prompts and test cases for your chatbots and safety filters, which a model that refuses won't write for you.

  • Fiction and roleplay

    Dark and mature stories and characters, written without the model stepping out of the story.

  • Long documents

    Reports, logs and transcripts up to 200K tokens, read and answered in one request.

Get started

Call GLM-4.7 Flash Heretic in three steps.

No waitlist and no new SDK. If your code already talks to OpenAI, it already talks to Zurelay.

  1. 1

    Create a key

    Sign up, add credit and create an API key. Set a monthly budget or a rate limit per key if you like.

    Get your API key
  2. 2

    Point your SDK at Zurelay

    Change the base URL. Everything else in your code stays the same.

    https://api.zurelay.com/v1
  3. 3

    Use glm-4.7-flash-heretic

    Send chat completions as usual. Streaming, tool calls and usage reporting work as you expect.

    glm-4.7-flash-heretic

Works with the tools you already use

Compare

Similar models. All on the same key.

62%off

GLM-5.3 Flash

$0.058 / $0.19 per 1M tokens

Pick GLM-5.3 Flash, Z.ai's own current low-cost GLM, for coding and agent work that doesn't need refusals removed.

GLM-5.3 Flash API
2%off

Abliterated Large V2

$2.93 / $4.88 per 1M tokens

Pick Abliterated Large V2 for a much larger model with its refusals removed and a 1M-token context window, at a higher price per token.

Abliterated Large V2 API
4%off

Gemma 4 Uncensored

$0.155 / $0.48 per 1M tokens

Pick Gemma 4 Uncensored for an uncensored model that reads an image and holds 256K tokens of context.

Gemma 4 Uncensored API

FAQ

GLM-4.7 Flash Heretic API questions.

Anything else? Ask our team and we’ll answer by email.

How much does the GLM-4.7 Flash Heretic API cost?

On Zurelay, glm-4.7-flash-heretic costs $0.061 per 1M input tokens and $0.35 per 1M output tokens, with cached input at $0.0305. The list price is $0.07 input and $0.40 output per 1M, so you save 13%. Reasoning tokens bill as output, one rate applies at every prompt length, and failed requests are never billed.

What is Heretic?

Heretic is an open-source tool that removes a model's refusals automatically. It compares the model's internal activations on prompts it refuses and prompts it answers, finds the direction linked to refusing and edits the weights to suppress it, tuning the edit so the model changes as little as possible otherwise. GLM-4.7 Flash Heretic is GLM-4.7 Flash after that edit, published by Olafangensan.

Is GLM-4.7 Flash Heretic made by Z.ai?

No. Z.ai made GLM-4.7 Flash and released its weights; GLM-4.7 Flash Heretic is a community edit of those weights, not made or endorsed by Z.ai. Asked what it is, it names itself GLM-4.7 Flash Heretic and credits Z.ai, whose model it's built on. For Z.ai's own models with their safety tuning intact, use GLM-5.3 Flash or GLM-5.3.

Do Zurelay's Terms of Service still apply?

Yes. Removing a model's refusals changes what it will write, not the rules. Zurelay's Terms of Service (https://zurelay.com/legal/terms) apply to everything sent to GLM-4.7 Flash Heretic, including the acceptable use rules in section 7, and you are responsible for how you use its output, as with any model on Zurelay.

How do I call the GLM-4.7 Flash Heretic API with the OpenAI SDK?

Install the official openai package, set base_url to https://api.zurelay.com/v1 and use your Zurelay API key. Then send a Chat Completions request with model set to "glm-4.7-flash-heretic". Zurelay also accepts olafangensan-glm-4.7-flash-heretic.

How do GLM-4.7 Flash Heretic reasoning levels work?

It reasons at low by default. Set reasoning_effort to none, low, medium or high. none switches reasoning off for a fast, direct answer; high spends the most thinking on hard problems, more than twice as long as low in our tests. The reasoning comes back in reasoning_content and bills as output.

What is the GLM-4.7 Flash Heretic context window?

It holds 200,000 tokens of context and writes up to 24,000 output tokens; a request with a higher max_tokens is refused. Reasoning and the answer share that budget. In our tests it found one fact buried in a 168K-token prompt in 44 seconds.

Does GLM-4.7 Flash Heretic support tool calling, images and JSON output?

It calls tools on its own with tool_choice auto, several at once and while streaming, reads tool results, and follows "required" and a tool forced by name. For JSON, send response_format with json_object or json_schema. It takes text only: a request with images or files is turned away with a clear error before anything is billed. logprobs isn't supported.

More models on Zurelay

All models

Stop paying list price.
Start saving today.

Create a key in seconds, point the SDK you already use at Zurelay, and every request costs up to 90% less from the first token.

Get your API keyTalk to an engineer