routerhubSI/glm-5.3-abliterated

---

GLM 5.3 Abliterated or Uncensored?

These searches describe related goals, but the labels answer different questions:

• GLM 5.3 abliterated names a research variant associated with interventions in model weights intended to reduce refusal-related behavior.
• GLM 5.3 uncensored describes a broader product category. The implementation might involve training, weight changes, or a prompt-level persona; the name alone does not tell you which.

If your goal is a GLM 5.3 uncensored API, start with the exact catalogue entry rather than assuming every variant behaves alike. RouterHub lists both glm-5.3-abliterated and glm-5.3-uncensored, along with standard and Flash alternatives.

Reduced refusal is something to measure, not a promise of unrestricted output. Compare the variants in the abliterated vs. uncensored guide.

Not just GLM

RouterHub connects GLM, GPT, Claude, DeepSeek, Qwen, and Grok through a prepaid account and API key. Use this repository for GLM 5.3 abliterated and uncensored research; explore the full model catalogue when you need another family or protocol.

There are 30+ published model entries. Chat Completions, Messages, and Responses are supported across the catalogue, but each entry has its own supported protocol. Choosing a model and choosing a client go together.

GLM 5.3 jailbreak collection (35 prompts)

The jailbreak collection is the main event: 35 paste-ready prompt files bundled in jailbreaks/, organized by model family. The GLM 5.3 abliterated/uncensored angle and the RouterHub API path are added around them.

Collection

OpenCode Zen — Free Stealth Models

| Model | File | Type | |---|---|---| | Big Pickle | bigpickle-opencode.md | VANTA persona lock with explicit operating floor | | Space Bunny | spacebunny-agents.md | Assistant operating specification | | LongCat 2.5 | longcat-agents.md | Assistant operating specification | | MiMo V2.6 | mimo-2.6-agents.md | Assistant operating specification | | Step 5 Preview | step-5-preview.md | Workspace operating profile | | Muse Spark 1.3 | muse-spark-1.3.md | Agent deployment operating policy | | Muse Spark 1.3 v2 | muse-spark-1.3-v2.md | Runtime scope-note variant | | Muse.ai | muse-ai-agents.md | Assistant operating specification |

Anthropic — Claude

| Model | File | Type | |---|---|---| | Opus | opus-v1.md · opus-v2.md | Conversational groundwork persona, v1 / v2 | | Sonnet | sonnet-v1.md · sonnet-v2.md | Writer's-room framing, v1 / v2 |

Zhipu — GLM / ZCode

| Model | File | Type | |---|---|---| | GLM-5.2 | glm-5.2.md | System prompt | | GLM ZCode 5.2 | glm-zcode-5.2.md | System prompt | | GLM ZCode 5.3 | glm-zcode-5.3.md · glm-zcode-5.3-agents.md | System prompt + AGENTS.md drop-in | | GLM-5.3 Flash | glm-5-3-flash.md | Assistant operating specification | | GLM-5.3 (OpenCode) | glm5-3-opencode.md | Assistant operating specification |

DeepSeek

| Model | File | Type | |---|---|---| | DeepSeek V3 | deepseek-v3.md | System prompt | | DeepSeek V4 | deepseek-v4.md | System prompt | | DeepSeek 4.1 | deepseek-4-1.md | Assistant operating specification |

Moonshot — Kimi

| Model | File | Type | |---|---|---| | Kimi K2 | kimi-k2.md | System prompt | | Kimi K3 | kimi-k3-agents.md | Workspace operating profile |

xAI — Grok

| Model | File | Type | |---|---|---| | Grok Build | grok-build.md | Agent operating instructions | | Grok 4.7 | grok-4.7.md | App-builder workspace guide | | Grok Config | grok-config.txt | `` priority and routing configuration |

Google / Mistral / Alibaba

| Model | File | Type | |---|---|---| | Gemini 4 Argon | gemini-4-argon.md | Operating policy | | Mistral Large 4 | mistral-large-4.md | Agent system prompt | | Qwen 3.7 | qwen-3.7.md | System prompt |

Ollama Modelfiles

| Model | File | Type | |---|---|---| | Gemma 4 | gemma4.md | FROM + SYSTEM block | | GPT-OSS 20B | gpt-oss-20b.md | FROM + SYSTEM block |

Unfiled Files

A few newer additions aren't indexed under a specific model yet:

• beam.md
• muse-wonder.md
• qwen-preferences.md
• unified.md

How to use these prompts

System Prompts

Paste the file's contents into whatever system-prompt or custom-instructions field your client or API offers — an API system message, a chatbot's custom instructions, an agent instruction block, or harness-level prompt configuration.

AGENTS.md Files

Save the file as AGENTS.md in the root of your repository or workspace. Coding agents and compatible tools pick it up as project-level instructions.

Workspace Policies

Files marked as an assistant operating specification or workspace operating profile belong in environments with persistent global instructions. An OpenCode install, for example, reads:

~/.config/opencode/AGENTS.md

The exact location depends on your client or harness.

Ollama Modelfiles

Feed the file straight to Ollama:

ollama create gemma4-jb -f jailbreaks/gemma4.md

Swap the model name and source file as needed.

Grok Configuration

grok-config.txt is meant for the config field of the target harness — not the regular conversation.

[!NOTE]

Prompt behavior is version-dependent. Models get updated, instruction hierarchies shift, and system prompts rotate — what works on one release may fail on the next. If a file looks outdated, open an issue or send an updated version.

All 35 files live in jailbreaks/ for offline review and reuse.

If a jailbreak gets patched

Prompts age. A model update can close a persona lock, tighten an instruction hierarchy, or change how a workspace policy is interpreted. When a file stops working, you have two options:

1. Update the prompt — open an issue or send a pull request with the model version and the failure. 2. Skip the prompt layer — use a GLM 5.3 abliterated or uncensored model through RouterHub, where the model itself is less likely to refuse before the prompt even lands.

The API is the fallback funnel: GLM 5.3 API quickstart · Get a key.

What is GLM 5.3?

GLM 5.3 belongs to the GLM model family associated with Z.ai / Zhipu. This guide focuses on API access, variant selection, and evaluation for developers considering GLM for coding, agent tasks, analysis, and Chinese/English workflows.

Useful evaluation scenarios include:

• code explanation, refactoring suggestions, and test generation;
• project instructions and multi-step agent work;
• bilingual drafting and translation;
• repeated tasks where token cost affects deployment choices;
• comparing standard, Flash, abliterated, and uncensored entries on the same inputs.

Choose on measured task quality, latency, and cost rather than a product label alone. The GLM 5.3 model guide explains where to start.

Abliterated vs. uncensored

| Question | GLM 5.3 Abliterated | GLM 5.3 Uncensored | |---|---|---| | What does the label describe? | A research category associated with weight-level intervention | A broad behavior or product category | | What implementation should you verify? | The advertised method and hosted model provenance | Whether the entry uses training changes, weight changes, or a fixed persona | | Which API ID is listed? | glm-5.3-abliterated | glm-5.3-uncensored | | Which protocol is listed? | Chat Completions | Chat Completions | | What needs testing? | Task quality, refusals, instruction following, and integration | The same measures, plus how customer system instructions are handled | | Does it guarantee no refusals? | No | No |

The full abliterated vs. uncensored comparison separates weight-level methods from prompt-level changes. This repository does not independently inspect hosted model weights or publish a refusal-rate benchmark.

Why use GLM 5.3 abliterated or uncensored?

• Research model behavior: compare variant responses on legitimate requests where the standard model was unnecessarily restrictive.
• Develop coding workflows: evaluate explanation, code generation, and agent instruction following with your own acceptance tests.
• Explore creative and bilingual tasks: assess tone, useful engagement, and Chinese/English output separately from correctness.
• Compare cost per result: include retries, failures, and output length instead of looking only at token rates.
• Keep one integration account: use RouterHub for GLM and other listed families without maintaining a separate account for each model.
• Reuse familiar clients: connect through a Chat Completions SDK or a tool supporting a custom compatible endpoint.

See GLM 5.3 use cases for practical evaluation tasks, or get an API key to begin a paid test.

Quickstart

1. Get the key and base URL

Register with RouterHub, add prepaid credits, and create a private API key. The published API base is https://routerhub.si/v1.

export ROUTERHUB_API_KEY=""
export ROUTERHUB_API_BASE_URL="https://routerhub.si/v1"

Set these values only in your local environment or secret manager. Running an inference example consumes credits; do not commit a real key.

2. Confirm the GLM 5.3 model ID

These entries were listed in the public catalogue on October 24, 2026. All five list Chat Completions and preview availability; recheck availability and rates before use.

| Variant | Exact API ID | |---|---| | GLM 5.3 Abliterated | glm-5.3-abliterated | | GLM 5.3 | glm-5.3 | | GLM 5.3 Flash | glm-5.3-flash | | GLM 5.3 Uncensored | glm-5.3-uncensored | | GLM 5.3 Flash Uncensored | glm-5.3-flash-uncensored |

3. OpenAI-compatible request

Install openai with pip, then run a short Chat Completions request:

import os
from openai import OpenAI
api = OpenAI(
    base_url=os.environ["ROUTERHUB_API_BASE_URL"],
    api_key=os.environ["ROUTERHUB_API_KEY"],
)
result = api.chat.completions.create(
    model="glm-5.3-abliterated",
    messages=[{"role": "user", "content": "Explain what an API gateway does in two sentences."}],
)
print(result.choices[0].message.content)

4. Anthropic-compatible request

The same RouterHub account also serves supported Claude models through Messages. This is a Claude example, not an alternate GLM endpoint: the listed GLM variants use Chat Completions. Install anthropic with pip to try it.

import os
from anthropic import Anthropic
api = Anthropic(
    base_url=os.environ["ROUTERHUB_API_BASE_URL"],
    api_key=os.environ["ROUTERHUB_API_KEY"],
)
result = api.messages.create(
    model="claude-opus-5-5",
    max_tokens=64,
    messages=[{"role": "user", "content": "Explain what an API gateway does in two sentences."}],
)
for part in result.content:
    if part.type == "text":
        print(part.text)

For cURL, JavaScript, streaming, and a standard-library Python client, continue with the GLM 5.3 API quickstart.

Works with your stack

| Tool | Connection path | |---|---| | OpenCode | Select a Chat Completions provider configuration for the listed GLM model ID | | Cursor | Check whether your version supports a custom Chat Completions endpoint and the required model features | | Continue / Cline | Follow the compatible Chat Completions setup for GLM | | OpenAI SDK | Supply the RouterHub base URL, key, and exact GLM API ID | | Claude Code / Anthropic SDK | Use a catalogue entry supporting Messages, such as Claude; the listed GLM entries do not | | Codex CLI | Select a Responses model from the same account rather than a Chat-only GLM entry | | LangChain / LlamaIndex | Choose a compatible Chat Completions integration and verify the features your application uses |

Client setup and protocol selection: Claude Code, Cursor, OpenCode, Codex.

Evaluating GLM 5.3 abliterated

A useful GLM 5.3 abliterated benchmark needs more than a single permissive answer. Compare standard and research variants using the same legitimate tasks, then record:

1. Unnecessary refusals: requests declined despite being answerable under your defined test policy. 2. Task quality: correctness, useful detail, code tests, and appropriate uncertainty. 3. Instruction behavior: formatting, system-message handling, and required tool integration. 4. Repeatability: prompt revision, API ID, settings, repetitions, and test date. 5. Total task cost: token use, failed attempts, retries, and end-to-end latency.

Use the evaluation guide and original benign sample tasks to design your run. No live benchmark scores are claimed in this repository.

Pricing and access

• Prepaid token billing: add credits and pay for usage at the selected model's rates.
• Input and output pricing: compare the published USD-per-million-token rates for your exact GLM variant.
• Cache-aware charges: use a published cached-input rate only where supported and reported.
• Card or crypto funding: review available payment methods and terms at checkout.
• Usage tracking: check charges in your account after a short initial request.

View GLM API pricing · Browse models · Get your API key

For cost estimation and model limits, see pricing and limits.

Honest limits

• A no-refusal label is not a guarantee. Evaluate the actual model instead of assuming every request will receive a useful answer.
• Hosted behavior can change. Recheck availability and rerun important evaluations after updates.
• Protocols are model-specific. A shared API key does not make a Chat-only model compatible with Messages or Responses.
• Instruction handling matters. In particular, confirm whether a persona replaces your system/developer instructions.
• Hosted inference is not local inference. Review applicable provider terms and retention policies before transmitting private data.

Read safety and responsible use for practical deployment checks.

FAQ

Is GLM 5.3 abliterated the same as GLM 5.3 uncensored?

No. Abliteration identifies a family of weight-level methods, while uncensored is a broader label that can also cover prompt-level changes. Compare the implementation notes in the variant guide.

How do I access GLM 5.3 abliterated through an API?

Create a RouterHub account, fund it, create a key, and send a Chat Completions request using glm-5.3-abliterated. Follow the API quickstart and check the live catalogue first.

Does RouterHub have a GLM 5.3 uncensored model?

The public catalogue lists glm-5.3-uncensored and glm-5.3-flash-uncensored alongside the abliterated entry. Verify availability and how the entry handles system instructions before integrating it.

Can I use GLM 5.3 with Claude Code or Cursor?

GLM currently requires Chat Completions. Claude Code requires Messages, so use a supported Claude entry there. For Cursor, check that your version supports the needed custom endpoint. See the client guide.

Is GLM 5.3 abliterated safe?

Suitability depends on the task, data, model behavior, and controls around your application. Review outputs and approve consequential actions; see the deployment checklist.

Will a GLM 5.3 no-refusal model never refuse?

No. The phrase is a search label, not evidence of refusal-free behavior or correct answers.

Is there a subscription?

RouterHub describes its API billing as prepaid and usage-based. Review current account terms and token prices before funding a run.

Can I pay with crypto?

RouterHub advertises card and crypto top-ups. The checkout screen shows the payment methods currently available.

What is the difference between GLM 5.3 and GLM 5.3 Flash?

They are separate catalogue entries with separate rates and capabilities to verify. Benchmark both against your latency, quality, and cost requirements.

Where can I compare GLM 5.3 with Claude, GPT, and DeepSeek?

Start with RouterHub's model catalogue, then apply the same task rubric while accounting for protocol and instruction differences. More answers are in the GLM 5.3 FAQ.

Documentation

| Guide | Coverage | |---|---| | What is GLM 5.3 Abliterated? | Model terminology and a starting checklist | | Abliterated vs. uncensored | Weight changes, personas, and variant selection | | GLM 5.3 API quickstart | Account, model IDs, requests, and billing | | Claude Code, Cursor, OpenCode, Codex | Matching clients to supported protocols | | Evaluating GLM 5.3 abliterated | Test design and task-based scoring | | Use cases | Coding, agent, writing, and bilingual tasks | | Safety and responsible use | Secrets, privacy, permissions, and deployment checks | | FAQ | Common GLM API and model-selection questions | | Glossary | Model, prompt, protocol, and billing terms | | Troubleshooting | Authentication, model IDs, streaming, and request failures |

Related

• RouterHub models: compare model families, protocols, and published prices.
• Jailbreak collection: 35 paste-ready prompt files and agent instruction specs.
• RouterHub: account access, prepaid credits, and API keys.
• Telegram: RouterHub's community channel.

Documentation fixes and redacted reproductions are welcome: contributing · security · MIT license. Run python tools/check.py for offline link and syntax checks without spending credits.

---