Overview
Refuseless is a hosted inference service for refusal-removed open-weight models. Its OpenAI-compatible interface is designed to make the available model lineup accessible through a familiar API format.
The service highlights models aimed at agentic coding, cybersecurity, terminal work, and automation. Published benchmark results distinguish the models by task rather than presenting them as a single general-purpose option.
Model Lineup and Benchmarks
The listed lineup includes two DeepSWE entries alongside models associated with cybersecurity and exploit-solving tasks. Reported results include 66.9 and 63.4 for agentic coding, 84.5 for CyberGym, 105 of 130 exploits solved in ExploitGym, 84.3 for Terminal Bench 2.1, and 48.8 for Automation Bench.
These figures provide task-specific reference points for comparing the hosted models. They do not indicate that every model shares the same specialization or benchmark profile.
API and Context
Refuseless exposes hosted inference through an OpenAI-compatible API. The pricing information also identifies a one-million-token context, supporting workloads that need to process large amounts of text within a request context.
Pricing and Payment
Usage is priced per million tokens, with separate rates for input, cached input, and output. The published lineup shows input rates from $0.15 to $2, cache rates from $0.08 to $0.35, and output rates from $0.65 to $3, depending on the selected offering.
Payments are accepted through Visa, Mastercard, and cryptocurrency.