GLM-5.3
Abliterated.
A prepaid API for text conversations and agent workflows.
Model access through Oathbreaker
Oathbreaker AI, operated by Formacha LLC, provides this model through its Chat Completions endpoint at https://oathbreaker.ai/v1. Use the public model identifier oathbreaker-glm-5.3. The upstream provider is Refuseless, where the identifier is glm-5.3-abliterated.
Oathbreaker operates the API gateway, account console, and prepaid billing. It does not develop the underlying GLM model. Requests pass through our hosting infrastructure to the inference provider; review the data handling policy before sending sensitive information.
Current capabilities and limits
| Property | Support |
|---|---|
| Request format | OpenAI-compatible Chat Completions |
| Input / output | Text / text |
| Streaming | Server-sent events, including final usage |
| Function calling | Supported; up to 128 tool definitions |
| Context admission | 131,072 tokens shared by input, safety headroom, and requested output |
| Maximum output | 32,768 tokens, including reasoning |
| Maximum request size | 1,048,576 bytes |
| Reasoning effort | Low, high, or max |
Input, output, request size, and rate limits apply independently. Tool calls are requests for your application to execute; Oathbreaker does not run your tools. Image input, audio, embeddings, and strict JSON Schema enforcement are not supported. See the full API limits.
Token pricing
| Token type | Price |
|---|---|
| Input | $3.00 |
| Cached input | $0.30 |
| Output, including reasoning | $5.00 |
Cached input is billed at the cached rate only when the provider reports a cache hit. A repeated prompt does not guarantee a hit.
Add prepaid credit starting at $5. There is no recurring subscription or automatic refill. The gateway reserves credit for a request’s possible cost and settles confirmed usage. Review pricing and billing details.
What does “abliterated” mean here?
“Abliterated” is the upstream model variant’s label for modifications intended to reduce refusal behavior. It is not a guarantee that every request will be answered, that an answer will be correct, or that the model will complete a task reliably. Oathbreaker does not independently certify the supplier’s modification process.
Evaluate the model on your own workload, validate outputs, and keep human approval around consequential actions. The service’s acceptable-use terms still apply.
Connect your application
Use the Python and JavaScript quickstarts or follow the Hermes and agent setup guide. The API reference covers authentication, streaming, tool messages, errors, and retry behavior.
Create your account