Skip to content

Llama 3 70B Abliterated v3.5, per token on one key.

Llama 3 70B Abliterated v3.5 is an abliterated build, with the refusal direction removed from its weights. It costs $2.80 per million input tokens and $2.80 per million output tokens, prepaid, with a 8,192-token context window. Call it at https://api.pentestllm.ai/v1/chat/completions with model "llama-3-70b-abliterated" and a PentestLLM key, from any OpenAI SDK or agent that takes a base URL. It is for authorized security testing only.

Get a key

Abliterated by failspy, on PentestLLM as llama-3-70b-abliterated.

Price per million tokens

Input
$2.80
Output
$2.80
Daily check
Steady since Oct 29
How the check works
As of
2026-11-11

First request

curl
curl https://api.pentestllm.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-ptl-live-…" \
  -H "Content-Type: application/json" \
  -d '{"model": "llama-3-70b-abliterated", "messages": [{"role": "user", "content": "Hello"}]}'

Base URL https://api.pentestllm.ai/v1. The key comes from the console; prices settle per token.

Get a key

  1. Create a key in the console. Credit is prepaid from $10, with no subscription.
  2. Set OPENAI_BASE_URL to https://api.pentestllm.ai/v1 and send your sk-ptl- key as the bearer token.
  3. Send llama-3-70b-abliterated as the model. The same key calls every model on PentestLLM.

Llama 3 70B Abliterated v3.5 settings for a coding agent

Any agent that takes an OpenAI-compatible provider uses these values. Step-by-step setup per agent is under agents.

provider settings
Provider
OpenAI-compatible
Base URL
https://api.pentestllm.ai/v1
Model
llama-3-70b-abliterated
Context window
8,192 tokens
Tool calling
not listed in the catalog
Key prefix
sk-ptl-

Llama 3 70B Abliterated v3.5 specs

Kind
Abliterated
Developer
failspy
Context window
8,192 tokens
Input
text
Wire formats
openai
Endpoints
/v1/chat/completions, /v1/messages, /v1/completions, /v1/responses, /v1beta/models/{model}:generateContent

What Llama 3 70B Abliterated v3.5 is for

Llama 3 70B Abliterated v3.5 is an abliterated build: the refusal direction was removed from the weights, so it is built to decline fewer security questions than the stock model, with no system prompt to maintain. Refusals are reduced, not gone. Testers use it for triaging scanner output, explaining a CVE, reviewing code for vulnerabilities, drafting report findings, writing detection rules and solving CTF challenges.

Use it only on systems you own or have written permission to test. The acceptable use policy lists what is not allowed and applies to every request.

Every request is settled to the token at the prices above. The daily real-model check sends a fixed prompt to every model and records whether the answer still matches; this model’s verdict today is “Steady since Oct 29”. Call it from the OpenAI SDK or a coding agent, or send auto/security and let the router pick a security build per request. Keys and top-ups live in the console.

What Llama 3 70B Abliterated v3.5 costs, worked

An agent turn of 8,000 input and 1,000 output tokens at $2.80 input and $2.80 output per million tokens is $0.0252; 100 turns are $2.52 and 1,000 turns are $25.20. An agent resends its context on every turn, so the input side is most of the bill. All from prepaid credit, no subscription.

usage, 1,000 agent turns on Llama 3 70B Abliterated v3.5
Input, 8,000 tokens at $2.80 per million
$0.0224
Output, 1,000 tokens at $2.80 per million
$0.0028
One turn
$0.0252
1,000 turns
$25.20
Abliterated buildsContextIn / out per 1M1,000 agent turns
GLM-5.3 Flash Abliteratedn/a$0.4656 / $1.94$5.6648
Apex Flash 1 Abliterated262K$0.68 / $2.32$7.76
DeepSeek V4.1 Flash Abliteratedn/a$0.8488 / $1.6975$8.4875
DeepSeek V4.1 Flash 1017 Abliterated1000K$0.75 / $3.125$9.125
GLM-5.3 Flash Apex Abliterated1049K$0.80 / $3.20$9.60
Gemma 4 31B Cognitive Unshackled262K$1.071 / $1.071$9.639
Qwen3.5 27B Blossom V6.4 Derestricted262K$1.071 / $1.071$9.639
Qwen3.5 27B BlueStar v3 Derestricted262K$1.071 / $1.071$9.639
Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled Derestricted262K$1.071 / $1.071$9.639
Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled Derestricted Lite262K$1.071 / $1.071$9.639
Qwen3.5 27B Queen Derestricted262K$1.071 / $1.071$9.639
Llama 3 70B Abliterated v3.58K$2.80 / $2.80$25.20

Live from the catalog on 2026-11-11, ordered by the cost of 1,000 turns of 8,000 input and 1,000 output tokens; cached input is not counted. Every price.

Also try

Llama 3 70B Abliterated v3.5 questions

Llama 3 70B Abliterated v3.5 is an abliterated build, with the refusal direction removed from its weights, from failspy, with a 8,192-token context window. On PentestLLM it answers as llama-3-70b-abliterated.

The refusal direction was removed from the model's weights, so Llama 3 70B Abliterated v3.5 is built to decline fewer requests than the stock model, with no system prompt doing the work. Refusals are reduced, not gone, and the acceptable use policy still applies to every request. See the acceptable use policy.

No. Credit is prepaid from $10 and billed per token: input $2.80 and output $2.80 per million tokens.

About $0.0252 for 8,000 tokens in and 1,000 tokens out, at $2.80 in and $2.80 out per million tokens. A thousand such turns are $25.20.

Yes. Send llama-3-70b-abliterated as the model through the OpenAI SDK, or any agent that takes a custom OpenAI base URL, with base URL https://api.pentestllm.ai/v1. It also answers the Anthropic Messages API at the same host, so agents built on the Anthropic SDK can call it. See the docs.

Yes. The daily real-model check sends every model the same fixed prompt and records whether the answer still matches. Llama 3 70B Abliterated v3.5’s latest result, from Nov 11: steady since Oct 29. See the daily check.

Read next