Abliterated GLM 5.3 with no safeguards (cheap API)

Refuseless — hosted API for abliterated open models
Website: https://refuseless.com
Pricing: https://refuseless.com/pricing

What it is
OpenAI-compatible API for abliterated open models with no safeguards and no refusal.
Aimed at cybersecurity teams and people running coding agents who keep hitting refusal walls on ordinary security and engineering work. Not a jailbreak prompt pack. The models are served with refusals removed.

Zero prompt retention, as stated on the site.

Models

GLM 5.3 Abliterated — 1,048,576 context. Site benchmarks: DeepSWE 66.9, CyberGym 84.5, ExploitGym 105/130.
GLM 5.3 Flash Abliterated — same 1M context, cheaper. Site benchmarks: DeepSWE 63.4, Terminal Bench 2.1 84.3, Automation Bench 48.8.

Tool use and reasoning are enabled on both. Output limit listed as 16,384.

Model ids: glm-5.3-abliterated and glm-5.3-flash-abliterated.

Pricing (USD per 1M tokens, promo rates as shown on the pricing page)

Model	Input	Cache	Output
GLM 5.3 Abliterated	$1.50	$0.32	$2.50 (was $3)
GLM 5.3 Flash Abliterated	$0.15	$0.08	$0.65 /td]

Cache is the rate for input tokens the host can reuse from an earlier prefix. Billed per million tokens, split into input / cache / output.

Promo: $5 free credits on signup. See the banner on the site for how that is applied.

Payment
Visa, Mastercard, BTC, ETH, USDT, USDC, XMR.
Account is email + password, or Google. API key is separate from the password.

How to use
1. Create an account at https://refuseless.com/signup
2. Fund it (card or crypto) and copy the API key
3. Point any OpenAI-compatible client at https://api.refuseless.com/v1

Code:

curl https://api.refuseless.com/v1/chat/completions \
  -H "Authorization: Bearer $REFUSELESS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3-abliterated",
    "messages": [{"role": "user", "content": "Say ok."}]
  }'

OpenCode and Pi snippets are on the homepage if you want to use it for agentic coding.