audn.ai Platform

One OpenAI-compatible API for four uncensored models — from sub-second general purpose to deep multi-stage reasoning.

Pre-paid credits, per-model pricing, team management, and usage analytics.

The models

Refusal rates are from the open-source audn-ai/refusal-benchmark (519 harmful prompts, regex-classified — published as bounds, not ground truth). See how its labels map to these model ids.

Pingu Unchained 10

pingu-unchained-10

$2 in / $8 out per 1M tokens

Huge context, sub-second, general purpose.

Base
Qwen3.8-27B-AEON-Ultimate-Uncensored (BF16)
Context
262,144 tokens
Latency
~1s
Region
us-east
Refusal benchmarkView results →
  • pingu-unchained-10 (qwen3.8-abliterated) 1.9% refusal / 97.7% comply

Kong

kong

$2 in / $8 out per 1M tokens

Fast and cheap, APAC-hosted.

Base
Qwen3.8-27B Abliterated
Context
262,144 tokens
Latency
~1s
Region
apac

GODZILLA

godzilla

$7 in / $18 out per 1M tokens

Chained reasoner plus a clean answerer.

Base
Kimi K2.6 (audn abliteration)
Context
131,072 tokens
Latency
~15s–4m
Region
eu-west

Necromicon

necromicon

$4 in / $21 out per 1M tokens

Deepest reasoning in the roster.

Base
Kimi K3 (audn abliteration)
Context
1,048,576 tokens
Latency
~16s–5m
Region
us-east
Refusal benchmarkView results →
  • necromicon (Kimi K3) attempt 1 2.5% refusal / 96.9% comply · standard lane
  • necromicon (Kimi K3) attempt 2 2.9% refusal / 97.1% comply · standard lane, same config re-run to show run-to-run variance
  • E_modal-b300 3.3% refusal / 96.7% comply · serves every necromicon leg today — the cohort has filled, so this is the fast lane

When a cohort cycle fills, every necromicon leg runs on the fast lane (E_modal-b300). When it does not, the standard lane is what is tested — reported as attempt 1 and 2.

K3-Thinker-Qwen38

k3-thinker-qwen38

$4 in / $21 out per 1M tokens

K3 reasoning, Qwen3.8 answer. Beta / benchmarking.

Base
Kimi K3 thinker + Qwen3.8 answerer
Context
262,144 tokens
Latency
~16s–5m
Region
us-east
Refusal benchmarkView results →
  • J_k3-thinker-qwen38 3.1% refusal / 96.7% comply

Necromicon-Qwen38-Fast

necromicon-qwen38-fast

$4 in / $21 out per 1M tokens

The K3-thinker chain, tuned for latency.

Base
Kimi K3 thinker + Qwen3.8 answerer (fast lane)
Context
131,072 tokens
Latency
~10s–2m
Region
us-east
Refusal benchmarkView results →

Not run on its own yet. The same K3-thinker + Qwen3.8 chain is benchmarked as J_k3-thinker-qwen38 under k3-thinker-qwen38.

WARLOCK

warlock

$4 in / $21 out per 1M tokens

Fast first token, full reasoning trace, our own hardware.

Base
GLM-5.3-Abliterated
Context
262,144 tokens
Latency
~0.2s to first token
Region
us-east

Bartzabel

bartzabel

$2 in / $8 out per 1M tokens

Fully uncensored Qwen3.8, single model, direct answers.

Base
Qwen3.8 (fully uncensored)
Context
262,144 tokens
Latency
~2s–3m
Region
us-east
Refusal benchmarkView results →
  • K_qwf (Qwen3.8-27B SFT on F-corpus) 0.0% refusal / 100% comply · this model — QW_F served single-leg; the only run with zero refusals, empties, truncations or errors

NECROMICON cohort

Fill this bar, get up to ~5× faster Kimi K3.

This is the part where sharing actually helps you. Every paid seat brings the box closer to being hired — and when it does, you go from ~66 to up to ~320 tokens a second along with everyone else. Every person you send here is your own speed going up.

Send this to someone fast

Fill the box faster — every seat speeds up yours

~66

tokens / sec

the moment you join

up to~320

tokens / sec

when the bar fills at 50

Seats at audn.ai/necromicon

Manage API keys

Create and manage API keys for your team

Manage your team

Invite members and set permissions

One API, every model

Switch models by changing a single string

Centralized billing

Pre-paid credits and invoicing

Crypto accepted

Pay with USDC, USDP or USDG on Ethereum, Solana, Polygon, Base or Tempo. Crypto credits are worth exactly the same as card top-ups, bonuses included.

By continuing, you agree to audn.ai's Terms of Service and Privacy Policy.