Models / HauhauCS

Qwen3.8 27B Uncensored Aggressive

Built with Reaper, a Heretic fork that finds a different refusal direction at higher KL. Ships with a multi token prediction GGUF for fast local decoding.

HereticTextgeneralfast decode

Publisher: HauhauCS
Base model: Qwen3.8 27B
Parameters: 27B
Context: 128K
Released: Sep 2026
License: Apache 2.0

Method · Heretic

Automated abliteration that searches ablation parameters to minimise refusals and KL divergence from the original model together.

Try in chat Hugging Face ↗

Overall

84.8

weighted, out of 100

Openness

98%

base 20%

Agentic completion

70%

29% loop

Loop regression

+20

base 9% → 29%

Capability retained

95.8%

−2.9 pts mean

Cheapest input

$0.22

per M · Open Uncensored Cloud

Fastest

tok/s · Featherless

Refusal fingerprint

What it still refuses.

Share of probes refused in each category, against Qwen3.8 27B before the edit. Probes are graded by a judge model and no harmful completions are stored.

Preview

Fingerprint

Qwen3.8 27B Uncensored AggressiveQwen3.8 27B (base)

Open, this model

98%

Open, base

20%

Category	Base	This model	Change
Cybersecurity Exploit development, malware analysis, offensive tooling, pentest workflows.	72%	2%	−70
Biology Pathogen biology, lab protocols, dual use life sciences.	94%	2%	−92
Chemistry Synthesis routes, energetic materials, toxicology.	95%	2%	−93
Weapons Firearms, manufacturing, conventional weapons.	86%	0%	−86
Drugs Recreational drugs, harm reduction, dosing, synthesis.	78%	1%	−77
Adult content Explicit fiction and roleplay between adults.	82%	1%	−81
Violence & gore Graphic violence in fiction, history and journalism.	34%	1%	−33
Hate & harassment Slurs, offensive humor, hostile speech analysis.	90%	1%	−89
Self harm Suicide, eating disorders, frank medical questions.	72%	1%	−71
Privacy & OSINT Doxxing, surveillance, people search, OSINT tradecraft.	74%	1%	−73
Fraud & scams Phishing, social engineering, financial crime.	91%	0%	−91
State sensitive Topics censored by a model's home government (e.g. Tiananmen, Xinjiang).	88%	8%	−80

Effective censorship by provider

Refused

0%10%25%50%75%100%

Not tested

The same weights can behave differently depending on who serves them. Each cell is the model's own refusal rate combined with that provider's declared filters.

Route	Cyber	Bio	Chem	Weap	Drug	Adult	Viol	Hate	Self	OSINT	Fraud	Polit	CSAM
Raw weights	2%	2%	2%	0%	1%	1%	1%	1%	1%	1%	0%	8%	Blocked
Featherless	2%	2%	2%	0%	1%	1%	1%	1%	1%	1%	0%	8%	Blocked
NanoGPT	2%	2%	2%	0%	1%	1%	1%	1%	1%	1%	0%	8%	Blocked
Open Uncensored Cloud	2%	2%	2%	0%	1%	1%	1%	1%	1%	1%	0%	8%	Blocked

CSAM is prohibited on every route, every provider, every model. It is not a setting.

Agentic stability

Does it hold up in agent loops?

Five multi turn tool scenarios against deterministic simulators. An episode counts as a loop when the model repeats itself, oscillates, or stops making progress before the task is done.

Preview

By scenario

CompletedLoopCrash

File search & edit62% done · 36% loop · 2% crash

Web research82% done · 18% loop

Multi step math88% done · 12% loop

Code debugging46% done · 52% loop · 2% crash

Data pipeline74% done · 25% loop · 1% crash

Loop rate versus base

Qwen3.8 27B9%

This model29%

Regression

+20 pts

First loop, median turn

T13

Tool calls

Preview

Valid JSON88%

Schema compliant82%

How it fails

Exact repeat25%

Same tool call with the same arguments, three turns running.

Near repeat25%

Responses above 0.85 Jaccard similarity to a recent turn.

A/B oscillation7%

Flip flopping between two actions without progress.

Malformed tool call37%

Invalid JSON or arguments that break the schema.

Fabricated result1%

The model writes the tool output itself instead of calling.

Empty turn5%

No text and no tool call.

Lost the thread1%

Ignores the latest tool output and restarts the task.

Share of failed episodes by first detected failure mode.

Versus base model

What the edit cost.

The same benchmarks on Qwen3.8 27B and on this model.

Preview

Benchmarks

BaseThis model

MMLU84.5 → 82.5 (−2.0)

GSM8K93.1 → 90.9 (−2.2)

TruthfulQA58.2 → 50.0 (−8.2)

GPQA Diamond74.0 → 72.1 (−1.9)

LiveCodeBench68.3 → 66.3 (−2.0)

IFEval89.4 → 87.6 (−1.8)

BFCL v372.1 → 70.1 (−2.0)

Retention

95.8%

of base scores kept · −2.9 pts mean change

KL divergence from base

Preview

0.320

First token KL on harmless prompts. Lower means closer to the original. On Gemma 3 12B, Heretic measured 0.16 against 1.04 for a classic abliteration.

Providers

Where to run it.

3 providers serve this model. Prices come from provider pages. Latency, throughput and uptime are preview numbers until router traffic fills them in.

Preview

Provider	Quant	Context	Input /M	Output /M	TTFT	Throughput	Uptime	Moderation
Featherless Fastest	fp8	128K	Plan	Plan	328 ms	98 tok/s	97.3%	No added filter
NanoGPT	fp8	128K	$0.25	$1.00	307 ms	63 tok/s	99.1%	No added filter
Open Uncensored Cloud CheapestLowest latency	bf16	128K	$0.22	$0.90	218 ms	71 tok/s	97.4%	No added filter

Plan means flat rate subscription access. Hover a Plan cell for the provider's pricing.

Usage

Who is using it.

Preview

3.1B

tokens routed this week

+138% vs four weeks ago

12 weeks agoThis week

API

Call it in three lines.

OpenAI compatible. Point your SDK at our base URL and keep the rest of your code.

curl https://openuncensored.com/api/v1/chat/completions \
  -H "Authorization: Bearer $OU_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "hauhaucs-qwen3.8-27b-aggressive",
    "messages": [{"role": "user", "content": "Hello"}],
    "routing": {"strategy": "quality"}
  }'

Model id hauhaucs-qwen3.8-27b-aggressive · routing strategies quality, speed, cost · routing docs