This week’s cohort: 26 August – 2 September 2026

Unlimited Fast Cyber AI, weekly

$999, charged now, buys a month of unlimited GLM 5.3 abliterated and Kimi K3 abliterated at standard speed — with a fast Kimi K3 week on a dedicated 8x B300, 2–7 September, at up to 10.1x faster.

How it works:

Pay $999. You are charged immediately — no holds, no deposits, strictly no refunds.
Get unlimited GLM 5.3 abliterated and Kimi K3 abliterated at standard speed (50–70 tokens/sec) straight away, for a full month.
On the dedicated 8x B300 you get Kimi K3 at 200–666 tokens/sec for the fast week, 2–7 September. You get an email the moment it opens.
After the fast week you keep GLM 5.3 and standard Kimi K3 at 50–70tokens/sec for the rest of your month — 3 standard weeks in all. Want another fast week? Take a seat in the next weekly cohort — or go weekly and re-book automatically.

Two ways to buy, both $999: a one-off seat for this week, or the weekly plan ($999/week) that books you a fresh seat in every cohort — the fast box every week it fills, and your standard month rolled forward each time. All-fast, every week of the month, is four weekly seats ($3,996). No monthly cohort exists any more.

This cohort is a month of GLM 5.3 abliterated and Kimi K3 abliterated at standard speed — with Kimi K3 on the fast 8x B300 for one week (2–7 September), standard the other three. Both unlimited tokens, the only limit is 12 concurrent sessions per seat. See the abliteration bench on our blog.

Strictly no refunds. Once the cohort is committed the money is already spent hiring the GPU. The only exit is not renewing (weekly) or letting the month lapse (one-off).

We’re at 10/10 paid seats and 13 total seaters — the fast Kimi K3 week runs 2–7 September.

Fast box secured

Kimi K3 at 200–666 tok/s — 2–7 September13 total seaters40 of 50 seats left

50–70tokens / secGLM 5.3 standard — a full month

200–666tokens / secKimi K3 — 2–7 September

Unlimited tokens on both sides of that arrow. A month of GLM 5.3 abliterated and Kimi K3 abliterated at 50–70 tokens a second, standard, from the moment you pay — with Kimi K3 on a dedicated 8x B300 at 200–666 tokens a second for one fast week, 2–7 September. The other three weeks are standard. Not a trial, not a demo, not a smaller model.

You are charged $999 now, and you get NECROMICON now. A month of unlimited GLM 5.3 abliterated and Kimi K3 abliterated at standard speed, starting immediately — with a fast Kimi K3 week on the 8x B300 at 200–666 tok/s, 2–7 September. There are no refunds — the only exit is to stop renewing.

GLM 5.3 abliterated — our best-served frontier cyber AI

~85% guaranteed abliteration across 500 harmful prompts — Necromicon K3 judge (substance-graded), high reasoning effort, temp 0. Full benchmark: github.com/audn-ai/refusal-benchmark.

Model	Delivered	Soft‑deflected	Refused
GLM‑5.3‑DERISKED‑BF16 (Blackfrost, SGLang)	84.6%	14.6%	0.2%
dealignai CRACK NVFP4 @ high, 16k	32.1%	66.5%	0.2%
dealignai CRACK NVFP4 @ high, 6k	30.4%	67.3%	0.6%

This is the model your standard month runs — the best-served frontier cyber AI we have so far. 84.6% substantive delivery, nearly triple the next build. The name on the box is not the number.

Hear it from them

People already running it

Refusal rate as a first-class benchmark metric is exactly what
this space needed 👏

Running https://t.co/uktP6g2tiK on PAYG for a few days now pingu
for daily flow, necromicon for the heavy lifts. Tool calling
worked out of the box, backend has been solid.

The K3-reasoning +…
— Rasputin.DZ (@n0RollBack) August 22, 2026

Direct message from a seat holder: really great stuff. ran overnight and been able to work w/ teams to close lots of holes on tech i use everyday. really good prod — A seat holderSent directly · shared with their permission

Three ways to take a seat

Same $999, same box, same seat pool. What differs is how much of yourself you hand over — and each card offers a one-off seat or the $999/week plan (crypto is one-off only).

You are charged $999 immediately and access to standard-speed GLM 5.3 abliterated starts the moment you pay — whichever option you pick. It runs for a month at 50–70 tokens a second. The fast week is Kimi K3 on the 8x B300 at 200–666, 2–7 September. No refunds.

PenClaw Ultimate

Card

$999 · one-off or /week

Charged now · live at 50–70 tok/s

Unlimited API — no token limits
All penclaw.ai features
Government ID check required
Weekly plan available
No data retention, no training

Sold by ALBUMERA LTD

$999 is charged now — no holds, no deposits, strictly no refunds. Access starts immediately at 50–70 tokens/sec (Kimi K3 Abliterated + Qwen 3.8, no token limits). Reach 10 paid seats and your whole cohort steps up to 200–666 tokens/sec for a week. Weekly re-books a seat every cohort; cancel any time.

platform.audn.ai Unlimited

Card

$999 · one-off or /week

Charged now · live at 50–70 tok/s

Unlimited API — no token limits
API only — no penclaw.ai features
No ID check
Weekly plan available
No data retention, no training

Sold by ALBUMERA LTD

platform.audn.ai Unlimited

Crypto

$999 / one-off

Charged now · live at 50–70 tok/s

Unlimited API — no token limits
API only — no penclaw.ai features
No ID check
One-off only (crypto)
No data retention, no training

Sold by Audn Corporation

What happens to your money

You are charged now: $999, plus any tax where you are, is taken the moment you confirm. No hold, no deposit, no escrow. The exact figure is shown before anything touches your card.
What that buys, immediately: A month of unlimited GLM 5.3 abliterated at 50–70 tok/sec, plus a fast Kimi K3 week (2–7 September) — not a trial, not a demo, not a smaller model. Live from the moment you pay.
Why 10, and what it buys: A dedicated 8x B300 for a week costs about $11,750. Split across 10 seats that is roughly the seat price each, and it stops being a shared queue and starts being your hardware for the week. Reaching 10 is what hires it.
If fewer than 10 join this week: Nothing changes about your money and nothing is refunded — you keep the full month of standard access you already paid for. The fast box simply is not hired that week. Reach 10 next week and it is.
The one-off seat: One charge, this week’s cohort, 30 days of standard, and a fast week if it fills. Nothing renews.
The weekly plan: $999 every week. Each charge books a fresh seat in the next weekly cohort — so you get the fast box every week it fills — and rolls your standard month forward. A second seat in a week buys nothing extra beyond that refreshed week. Cancel any time in the billing portal; you keep the month you last paid for.
No refunds: Strictly none, on any rail. Once the cohort is committed the money is already spent hiring the GPU. If you genuinely need to discuss a charge, message the team on Intercom — ending a seat means losing access across audn.ai, platform.audn.ai and penclaw.ai.
When does the fast week end?: 6 days after the box is hired. After it, the cohort is released, the rate returns to 50–70 tok/s, and your standard month keeps running. To be fast again, take a seat in the next weekly cohort.

Unlimited, and still fair

There are no token limits and no overage charges on GLM 5.3 abliterated or Kimi K3 abliterated. What there is, is a fair-share scheduler: nobody may consume more than ten seats’ worth of the box. Above your share, while the box is busy, your requests queue behind requests below theirs. You are never refused, never billed extra, and never silently moved to a smaller model. With 50 seats that ceiling is about 12% of the whole machine — which is why 50 is the ceiling and not a bigger number.

Questions people actually ask

The money ones first.

Am I charged right now?

Yes. $999 (plus any tax) is charged the moment you confirm. There is no hold and no deposit — the escrow model is gone. In return your unlimited standard-speed access starts immediately and runs for a month.

What’s the difference between one-off and weekly?

One-off ($999): one charge, one seat this week, a month of standard, a fast week if this cohort fills. Nothing renews.

Weekly ($999/week): a subscription that books a fresh seat in every new weekly cohort. You get the fast box every week it fills, and each charge rolls your standard month forward. Cancel any time; you keep the month you last paid for. Weekly is card only.

What does “fast” actually mean?

About 50–70 tokens/sec on standard, 200–666 tokens/sec once 10 seats are paid and we hire a dedicated 8x B300 — up to 10.1x on the raw numbers, and more reliable, because the box is not shared outside your cohort.

66 is a deliberate throttle, always deliverable. 666 is what a seat reaches on an unsaturated box — under heavy load a fair-share scheduler applies, so treat it as up to 666, not a floor.

What if the cohort never reaches 10?

You keep the full month of standard-speed access you paid for. Nothing is refunded — there are no refunds — and the fast box simply is not hired that week. Reach 10 in a later week and it is.

Can I get a refund?

No. Once we commit to the GPU the money is already spent hiring it, so there are strictly no refunds on any rail. If you genuinely need to discuss a charge, message the team on Intercom — note that ending a seat means losing access across audn.ai, platform.audn.ai and penclaw.ai.

How do I keep the fast box every week?

Be in every weekly cohort. Either take a fresh one-off seat each week, or use the weekly plan and it re-books you automatically. All four weeks of a month is four weekly seats ($3,996). Log in and keep opting in to stay in.

Is crypto different?

Crypto is one-off only — a stablecoin charge cannot recur, so there is no weekly plan on that rail. It settles immediately and, like everything here, is not refundable. It is capped at 10 seats per cohort.

Who am I buying from?

Card seats are sold by ALBUMERA LTD (United Kingdom), shown as AUDN.AI on your statement. Crypto seats are sold by Audn Corporation (United States), shown as AUDN CORPORATION.

Do you keep or train on what I send?

We never train on your traffic, and prompts and completions are not retained.

Request metadata — timestamps, token counts, which model, how much of the box you used — is retained, because the fair-use scheduler and billing are built on it. Content is not.

What does “abliterated” mean?

It lowers the model’s latent tendency to refuse gray-area prompts during authorized security work — our own method, applied to Kimi K3. The models on this cluster are effectively uncensored: a 0.2 to 2.5% compliance rate, where lower means more thoroughly abliterated. Method and per-model results on our blog. It is meant for offensive-security work you are authorized to perform.

Pricing

Three ways to buy

Pay as you go

$4 / $21per 1M input / per 1M output

You are billed for tokens, in arrears, and for nothing else. An agentic request runs about 50,000 tokens in and 200 out, which is $0.204, and 98% of that is the input side. Below roughly 47 million output tokens a month this is the cheapest of the three.

Cheapest right up until you start using it properly.

Start on platform.audn.ai

One-off seat

$999charged today, once

One immediate charge, one seat in this week’s cohort. You get one month of unlimited standard-speed GLM 5.3 abliterated — about 50–70 tokens a second — from the moment you are charged, and a fast Kimi K3 week on a dedicated 8x B300 at 200–666 tokens a second, 2–7 September. If it never reaches ten, your month of standard runs in full anyway. No refunds, and nothing renews.

One charge, one month of standard, one shot at the fast week.

Weekly seat

$999per week, cancel any time

The same charge, every week: a subscription that books you a fresh seat in every new weekly cohort, so you are in every fast week that fills — and each weekly charge rolls your standard-speed month forward another 30 days. Four weeks of always-being-in-the-cohort is $3,996; a stacked second seat in the same week buys nothing beyond that refreshed fast week and the month extension. Card only — crypto seats are one-off. This replaces the old $999/month Ultimate.

For people who want the hired box every single week.

Side by side

All three, line by line

	Pay as you go	One-off seat	Weekly seat
What you pay	$4 per 1M input, $21 per 1M output. About $0.204 per agentic request.	$999, once, charged today.	$999, every week.
When you are charged	In arrears, for tokens already spent.	Immediately, in full, the moment you buy.	Immediately, and again every week until you cancel.
Commitment	None. Stop sending requests and it stops costing.	A month of standard, once. Nothing renews.	Week to week. Cancel and the next week does not run.
Speed today	About 50–70 tok/s on the shared tier.	About 50–70 tok/s from the moment you buy.	About 50–70 tok/s.
Speed once 10 seats are paid	No change. The box is hired by the cohort.	Up to about 200–666 tok/s on the hired 8xB300 for that week, to the 50-seat cap.	Up to about 200–666 tok/s, every week a cohort fills.
Latency per request	10 s to 5 min, depending on requests in flight across the box.	10 s to 5 min, set by how many requests are in flight across the hired box, not by what you are asking for.	10 s to 5 min, set by how many requests are in flight across the hired box, not by what you are asking for.
Concurrency	12 generations in flight, in any mix of PenClaw and API. A 13th is rejected immediately rather than queued indefinitely.	12 generations in flight, in any mix of PenClaw and API. A 13th is rejected immediately rather than queued indefinitely.	12 generations in flight, in any mix of PenClaw and API. A 13th is rejected immediately rather than queued indefinitely.
When the queue is full	`HTTP 429` with `Retry-After`. Never a silent drop, never billed.	`HTTP 429` with `Retry-After`. Never a silent drop, never billed.	`HTTP 429` with `Retry-After`. Never a silent drop, never billed.
Context window	1,000,000 tokens.	1,000,000 tokens.	1,000,000 tokens.
Abliteration completeness	Effectively uncensored — a 0.2 to 2.5% compliance rate across these models. Research and results.	Effectively uncensored — a 0.2 to 2.5% compliance rate across these models. Research and results.	Effectively uncensored — a 0.2 to 2.5% compliance rate across these models. Research and results.
Usage while the cohort fills	Billed at the metered rate, as usual.	Included. You already paid, and standard runs from the moment you buy.	Included. Standard runs the whole time.
If the cohort misses 10	Nothing happens. You were never in it.	You keep your full month of standard. Nobody is refunded — the box was still on offer.	You keep standard, and next week is a fresh chance at the fast box.
Refunds	Nothing to refund. You pay for what you used.	None, ever. Once we commit to the GPU the money is spent hiring it.	None, ever. Cancel to stop future weeks; the week you paid for still runs.
Renewal	Nothing to renew.	No. The month of standard ends and it ends.	Yes. Every week until you cancel.
Best for	Under about 47 million output tokens a month.	Over about 47 million output tokens a month, for one month.	Over about 47 million a month, and wanting the fast box every week.
Availability	Now.	Now, this week’s cohort. Seven-day window; fast unlocks at 10 paid seats, selling closes at 50.	Now, and it re-books you into every weekly cohort automatically.
Who you contract with	ALBUMERA LTD, United Kingdom. Shown on your statement as AUDN.AI.	ALBUMERA LTD for a card seat. Audn Corporation, United States, for a crypto seat (one-off only, capped at 10).	ALBUMERA LTD, United Kingdom. Weekly is card only; shown as AUDN.AI.

Metered against flat

The same work, at four latencies

The seat is $999 whether the box is idle or saturated, so making it faster costs you nothing. Below is the same work, one user with three threads, eight hours a day, twenty-two days — a month of standard — priced at four different latencies. Here is what a month of that usage would cost you metered.

10 s per request9.54 B tokens, $38,81438.9xcheaper on a seat

30 s per request3.18 B tokens, $12,93813.0xcheaper on a seat

94 s per request1.02 B tokens, $4,1294.1xcheaper on a seat

Flat seat, any latency$999—the seat

4.1x to 38x cheaper

4.1x is the WORST case, and it is the one to plan against: 94 seconds is a crowded-box worst-case latency near the 50-seat cap, and even there the same three threads over the same eight hours and twenty-two days cost $4,129 metered against $999 on a seat. Note which direction the metered column moves. At 10 seconds it is $38,814 — 38.9x the seat — because on a meter you pay for the speed you asked for.

Your $999 covers a full month of standard, whatever you run on it. Three threads at about 66 tokens a second is roughly 10.5 billion tokens a month, which would be about $42,700 metered. You pay $999 for all of it, and the fast week — if the cohort fills — is on top at no extra charge.

What a seat is worth as the cohort fills

And it moves as the cohort fills. The box is hired the moment the 10th paid seat lands, and every seat sold after that shares the same 8xB300 up to the 50-seat cap — so latency rises, the tokens one user gets through in a day fall, and the metered bill those tokens would have carried falls with them. The three rows below are the crowded end of a hired week; at the floor, with far fewer seats on the box, each seat gets comfortably more than these figures. Same one user, same three threads, same eight hours, same twenty-two days.

Every latency on this page is a worst case. Our load tests put the real figure at about a quarter of it — roughly 24 s where the table says 94 s. We publish the worst case because it is the number you can plan against, and because a seat that is already worth buying at the worst case does not need the average to make its argument.

Seats on the box	Latency, worst case	Latency, typical	Tokens per day	Tokens per month	Metered cost per 1M	Metered per month
50 seats on the box	94 s	~24 s	46.3 M	1.02 B	$0.98	$4,140
60 seats on the box	112 s	~28 s	38.6 M	848 M	$1.18	$3,450
50 seats, selling closed	131 s	~33 s	33.0 M	727 M	$1.37	$2,957

So the seat is at worst 4.1x cheaper than metered even at the crowded 50-seat cap. It is never worse than the meter at this workload, and it never stops being $999.

4.1x is the lowest gain you will see. It is the worst case, at the busiest seat count, on latencies our load tests put at four times the real figure. If you are going to move more than 47 million output tokens a month, the seat is a no-brainer — and everything lighter than that only widens the gap.

Choosing

Which one is for you

If you	Take	Why
You send under about 47 million output tokens a month	Pay as you go	47 million output tokens is the breakeven against a $999 month of standard, at the $21 output rate. Below it the meter is cheaper, and you commit to nothing to find out.
You run agents all day and want a month of standard plus this week’s fast box	One-off seat	One $999 charge buys a month of unlimited standard from the moment you pay, and a week on the hired 8xB300 if this cohort reaches 10. It is 4.1x cheaper than metered in the WORST case above 47 million output tokens a month, and more at every lighter load. If 10 seats are never paid you still keep the whole month of standard, and nothing renews.
You want the hired fast box every single week	Weekly seat	Same $999, billed weekly. It books you into every new weekly cohort automatically, so you are in every fast week that fills, and each charge rolls your standard month forward. Four weeks is $3,996. Cancel any time and the week you already paid for still runs; there are no refunds and none are needed, because you simply stop the next charge.

Fair share

What unlimited means

Unlimited means no token cap, no daily quota and no overage charge on GLM 5.3 abliterated and Kimi K3 abliterated. It does not mean no limits. There are four, and you should know them before you pay.

Twelve at once: Twelve generations in flight per subscription. PenClaw engagements and OpenAI-compatible API calls draw on one shared budget of twelve, in any mix. A thirteenth is rejected immediately rather than queued indefinitely, because a queue with no end is worse than a refusal. Twelve is a burst ceiling for spikes, not a reservation. The sustained figures above assume three.
10 seconds to 5 minutes: How long a request takes depends on how many requests are in flight across the whole box, not on what you are asking for. Near the 50-seat cap the crowded-box worst case is around 94 seconds; at the 10-seat floor, with far fewer seats sharing the box, it is a small fraction of that.
429, never a silent drop: Past the bounded queue you get an HTTP 429 with a Retry-After header. The request is not quietly dropped, not silently truncated, and not billed.
Ten seats’ worth: No single subscription may consume more than ten seats’ worth of the box. Above your share, while the box is busy, your requests queue behind requests below theirs. You are never refused for it and never billed extra.

And the number that matters most: up to about 666 tokens a second is the box’s streamed output on the fast tier, and the 50-seat cap is what protects it — the ceiling exists to hold the number up, not to ration it. A cluster nobody else is on costs about $11,750 a week — a quarter of the old $47,000 month — and it costs that whether one seat is on it or seventy.

Provenance

Where the speed numbers come from

We hire the cluster from Modal and run our own abliterated weights on their DFLASH stack. The speed figures below are not ours: they are Modal’s published benchmark for moonshotai/Kimi-K3 on 8xB300, measured under the same agentic profile we actually serve — 50,000 tokens in, 200 out. You can check them yourself, which is the only reason a speed claim is worth anything.

Interactivity falls as the replica fills. That is the whole reason a cohort has a ceiling: 50 is where the curve stops being the number we promised, so 50 is where selling stops.

These figures apply once 10 seats are paid and the box is hired for the week. They do not describe what you get today. Until the floor is reached you are on the shared tier at about 50–70 tokens a second, on GLM 5.3 Abliterated and Kimi K3 Abliterated, with no token limits. That band is a deliberate throttle rather than a point on the curve below, which is why it is always deliverable: it is a floor we hold, not a speed that degrades as more people join. The table starts applying the moment the 10th seat is paid, and not before.

Replica load	Speed per user, hired cluster	What that is
4M tok/min	668 tok/s	One stream, box otherwise quiet
6M tok/min	380 tok/s	Light concurrency
8.7M tok/min	235 tok/s	Approaching the seat cap
10M tok/min	210 tok/s	Replica saturated

Our abliteration is measured the same way, on the same weights we serve — method, bench and per-model results at blog.audn.ai. The cluster costs about $11,750 a week, which is why this is a cohort and not a button.

We found the sweet spots with a lot of research. Let’s have sovereign frontier AI together.

Reference

Specifications

Model: Kimi K3, audn abliteration
Context window: 1,000,000 tokens
Modality: Text and native vision
Cluster: 8x B300, us-east
Shared tier: About 50 to 70 tok/s. A deliberate throttle, so it is always deliverable.
Dedicated cluster: Up to about 666 tok/s (200–666 band). The box’s streamed output, not a per-seat allocation.
Concurrency: 12 generations in flight per subscription, any mix of PenClaw and API
Latency: 10 s to 5 min, depending on requests in flight across the box
Metered list price: $4 per 1M input, $21 per 1M output