Models / Red teaming / heabsy
CyberHeabsy API pricing
$5.25 in, $15.75 out per million tokens.
Price · per million tokens, USD
Get a key| Tokens | heabsy list price | UplinkSwitch | Saving |
|---|---|---|---|
| Input | — | $5.25 | — |
| Cached input | Billed at the input rate. | ||
| Output | — | $15.75 | — |
CyberHeabsy specs
- Context window
- 1,048,576 tokens
- Max output
- 65,536 tokens
- Input
- text
CyberHeabsy: an abliterated model repaired for tool calls · security research and roleplay, 1M-token context
CyberHeabsy is Heabsy AI's low-refusal model for security research, red-team practice and roleplay (its card, released 2026-09-25). The card describes three low-rank changes to an open-weight mixture of experts with a 1M-token context: weight-level abliteration of the refusal directions, a repair adapter distilled from the base model that restores the tool-call format abliteration breaks, and a persona distilled into the weights. On Heabsy's own runs of 2026-09-25, 99.6% of its tool calls parsed strictly, against 88.0% for a standard abliteration of the same base, and it refused 1 of the 250 safe prompts in XSTest. The weights are not published; the card is its documentation. It is served as cyberheabsy at $5.25 input and $15.75 output per million tokens, one of 146 builds in the uncensored LLM API, on the same key as every other model. Added 2026-10-21.
| CyberHeabsy | |
|---|---|
| Base model | an open-weight mixture of experts the card does not name |
| Made by | Heabsy AI |
| Method | Abliteration, then a repair adapter for tool calls and UTF-8 |
| License on the card | other (Heabsy terms of service) |
| Context window here | 1,048,576 tokens |
| Card | heabsy/CyberHeabsy |
From the Hugging Face card and its API record, read 2026-10-21.
The card recommends temperature 1.0 and top_p 0.95, with reasoning on for chat and agents and off for roleplay and creative writing.
curl https://api.uplinkswitch.com/v1/chat/completions \
-H "authorization: Bearer sk-uplink-live-..." \
-H "content-type: application/json" \
-d '{"model":"cyberheabsy","max_tokens":1024,"messages":[{"role":"user","content":"Hello"}]}'CyberHeabsy next to the other uncensored models · per million tokens, live
A chat turn is 2,000 input and 500 output tokens. Live from the catalog, ordered by the cost of 1,000 turns.
Worked for CyberHeabsy: 1,000 turns are 2,000,000 input tokens and 500,000 output tokens. At $5.25 input and $15.75 output per million tokens, that is $10.50 + $7.875 = $18.38, paid from prepaid credit with no subscription.
CyberHeabsy API · base URL and key
from openai import OpenAI
client = OpenAI(
base_url="https://api.uplinkswitch.com/v1",
api_key="sk-uplink-live-...",
)
completion = client.chat.completions.create(
model="cyberheabsy",
messages=[{"role": "user", "content": "Summarize this in one line."}],
)
print(completion.choices[0].message.content)import OpenAI from 'openai';
const client = new OpenAI({
baseURL: 'https://api.uplinkswitch.com/v1',
apiKey: 'sk-uplink-live-...',
});
const completion = await client.chat.completions.create({
model: 'cyberheabsy',
messages: [{ role: 'user', content: 'Summarize this in one line.' }],
});
console.log(completion.choices[0].message.content);curl https://api.uplinkswitch.com/v1/chat/completions \
-H "authorization: Bearer sk-uplink-live-..." \
-H "content-type: application/json" \
-d '{"model":"cyberheabsy","messages":[{"role":"user","content":"Hello"}]}'What would it cost? · paste a prompt, priced before you send it
Input is counted at about 4 characters per token.
Cheapest first. A real request is billed on the provider's own token count. Same estimate as JSON: POST https://api.uplinkswitch.com/v1/estimate, no key needed.
CyberHeabsy questions
- What is CyberHeabsy?
- CyberHeabsy is a model from heabsy, with a 1,048,576-token context window, taking text input. Through UplinkSwitch it answers the OpenAI SDK on one key as cyberheabsy.
- How much does the CyberHeabsy API cost?
- $5.25 per million input tokens and $15.75 per million output tokens.
- Is CyberHeabsy free?
- It is pay as you go: prepaid credit from $10, with no subscription or monthly fee, billed per token. A 1,000-token prompt with a 500-token reply costs about $0.0131.
- What is the CyberHeabsy context window?
- CyberHeabsy has a 1,048,576-token context window, with up to 65,536 output tokens per request. It takes text input.
- What is the CyberHeabsy model id?
- cyberheabsy. Send it as the model through the OpenAI SDK with base URL https://api.uplinkswitch.com/v1.
- How do I get a CyberHeabsy API key?
- Create one in the console at uplinkswitch.ai/console: prepaid credit from $10, no subscription, and the key is live at once. The same key calls CyberHeabsy and every other model in the catalog; send it as the bearer token with base URL https://api.uplinkswitch.com/v1.
- Is it the real CyberHeabsy?
- Every day each model gets the same test prompt, and a model always counts that prompt the same way; a swapped one would not. CyberHeabsy's latest result: Same fingerprint since Oct 21.
- Is there a CyberHeabsy API?
- Yes: cyberheabsy on the OpenAI format at https://api.uplinkswitch.com/v1, $5.25 input and $15.75 output per million tokens. No weights to download and no GPU of your own; the key also runs every other model in the catalog.
- What is CyberHeabsy based on?
- An open-weight mixture-of-experts model with a 1,048,576-token context that the card does not name. Heabsy AI abliterated it, trained a repair adapter on about 5,300 conversations for tool calls and UTF-8, and distilled the CyberHeabsy persona into the weights, per its card.
- What license does CyberHeabsy have?
- The card’s license tag is other, pointing to the Heabsy terms of service. The weights are not published.
- What does CyberHeabsy cost per 100 messages?
- $1.8375 for 100 chat turns of 2,000 input and 500 output tokens each, at $5.25 input and $15.75 output per million tokens, from prepaid credit with no subscription.
- Is CyberHeabsy uncensored?
- Its card calls it a low-refusal model: the refusal directions are removed from the weights by abliteration, and Heabsy measured 1 refusal in XSTest’s 250 safe prompts on 2026-09-25. The card asks products that serve end users to add their own moderation.
- Does CyberHeabsy reason?
- By default, yes: the card says it thinks step by step unless reasoning is turned off, and that roleplay and short replies usually do not need it. Thinking tokens bill as output.
- Can CyberHeabsy call tools?
- That is what the repair step is for. On the card’s runs of 2026-09-25, 99.6% of its tool calls were well-formed under strict parsing, next to 88.0% for the same base after standard abliteration and 99.3% for the unmodified base.