What is Refuseless?

Refuseless describes itself as a hosted inference API for open-weight language models. It presents an OpenAI-compatible chat-completions format, so applications built around compatible SDKs may be able to connect by changing the API base URL, model identifier and account API key. The service offers both modified “abliterated” models and at least one model using the provider’s description of original weights.

The word “abliterated” refers to a model modification intended to reduce certain refusal behaviors. It does not mean that outputs are reliable, safe, legal, or appropriate for every task. Review generated content and keep human oversight for consequential decisions.

Models and possible use cases

The provider’s current lineup includes GLM-5.3 Abliterated, GLM-5.3 Flash Abliterated, DeepSeek V4.1 Flash 1017 Abliterated, and a non-abliterated DeepSeek V4.1 Flash model. Model identifiers, available capabilities and benchmark claims can change. Some examples on the provider’s site refer to results for the original base model rather than independent Refuseless evaluations; do not treat those figures as a guarantee of performance from the hosted service.

For routine development, compare the candidate model on representative prompts, latency, output quality, tool-calling behavior and total cost. Use a test environment and non-sensitive inputs first. A model marketed for cyber-related work should only be used for systems and security testing that you own or are explicitly authorized to assess.

API compatibility and setup

The general integration pattern described by Refuseless is an OpenAI-style chat-completions request with a model ID and a bearer API key. However, its public homepage currently lists https://api.refuseless.com/v1, while the developer documentation lists https://api.refuseless.dev/v1. Because these official pages conflict, check the current documentation or your console for the correct base URL before configuring an SDK; this guide intentionally does not choose one.

Create an API key in your account and store it in a secret manager or environment variable. Do not put the key in browser-side code, a public repository, a prompt, or a client application that users can inspect. Restrict access where the provider allows it, monitor balance and usage, and rotate a key if it is exposed.

Token pricing and billing

Refuseless lists usage prices per one million tokens and separates input, cached input and output. The amount depends on the model. For example, the pricing page currently lists GLM-5.3 Abliterated at $2.00 per million input tokens and $3.00 per million output tokens, while the Flash version is listed at $0.23 input and $0.70 output. Promotional rates and all prices can change, so consult the live pricing page before budgeting.

The pricing page says usage is pay-as-you-go and billed from a prepaid balance. Estimate both prompt and response volume, account for retries and long contexts, and set internal usage monitoring before moving from experiments to a production workload. Payment options, deposits and refund rules may differ; read the current terms and payment screen before adding funds.

Privacy and data handling

Refuseless’s privacy policy says prompts and model outputs are not saved or used to train models, and says prompt content exists only temporarily in GPU processing cache. The Terms of Service similarly say prompts are not retained beyond what is needed to run the service. These are provider statements, not an independent audit. Before sending personal, confidential or regulated information, review the current policy, your own legal obligations, any applicable data-processing terms and the implications of using a third-party model service.

The privacy policy separately describes account, login, payment and referral-attribution data. “No prompt retention” therefore should not be read as “the provider collects no data.” Keep secrets and personal identifiers out of prompts unless there is a clear need and an appropriate basis.

Responsible use and limits

Refuseless’s terms prohibit using the service to break the law, attack or probe systems without permission, distribute malware or infringe others’ rights. Reduced model refusals do not waive those rules. Do not use the API to bypass access controls, facilitate harm or produce unauthorized exploit activity. For legitimate security work, define scope in writing, test only authorized assets and apply responsible disclosure practices.

Like other language models, outputs can be wrong, incomplete or fabricated. Validate code in a sandbox, inspect tool calls, and do not let an unchecked response trigger payments, account changes or production actions. Model availability and service uptime can also change under the provider’s terms.

Who may find it useful?

Refuseless may be worth evaluating for developers who want an API-style interface to the listed open-weight models and who can independently review model behavior, costs and data handling. It may be a poor fit when a project requires a verified compliance certification, a guaranteed service level, a particular data residency arrangement, or independently audited privacy claims that the user has not confirmed.

For comparison, see our guides to Freebuff for coding-agent offers and Manus for multi-step AI tasks.

Checks before connecting

  • Confirm the base URL and model ID in current provider documentation; the public homepage and docs presently differ on the endpoint hostname.
  • Compare current input, cached-input and output prices and estimate usage before funding the account.
  • Keep the API key server-side and monitor its use and balance.
  • Review privacy, retention, payment, refund and service-availability terms.
  • Use the API only for lawful tasks and systems you are authorized to test.
  • Evaluate outputs and generated code before relying on them.

Frequently asked questions

What is Refuseless?

Refuseless is a hosted API for open-weight language models. It uses an OpenAI-compatible request format and offers both abliterated and original-weight models.

Is Refuseless compatible with OpenAI SDKs?

The provider describes an OpenAI-compatible chat-completions API. Its public homepage and developer documentation currently show different base hostnames, so confirm the current endpoint in the official docs or account before configuring a client.

How is Refuseless priced?

The pricing page lists per-million-token rates for input, cached input and output, with different prices for each model. Check current rates and your balance before sending production or high-volume requests.

Does Refuseless retain prompts?

The provider’s privacy policy says prompts and outputs are not saved or used to train models, while terms say prompts are not retained beyond what is needed to run the service. Review the current policy and assess your own confidentiality obligations before sending sensitive data.

Does “abliterated” mean unrestricted or lawful for every use?

No. It describes a model modification intended to reduce refusals; it does not remove legal, security or service-policy responsibilities. Refuseless terms prohibit unlawful activity, unauthorized system probing or attacks, malware distribution and rights infringement.

Reviewed against publicly available Refuseless product, pricing, documentation, privacy and terms pages on 25 October 2026. Details may change.