Compare · prices checked 10 Oct 2026
A busy key is cheaper here.
abliteration.ai and Featherless charge by the token. We charge $499 or $899 and stop counting. Run the key like a product, all month, and the meter is the larger number.
The month below is 20 requests every minute, for 30 days, each call 2,000 tokens in and 1,000 out, with no cache hits. That is 864,000 calls, 1.728 billion input tokens, and 864 million output tokens. It sits inside our Standard cap of 60 requests a minute. It fits 4 requests in flight when a call finishes in about 12 seconds. Longer calls wait on that cap, and the other bill still follows whatever gets through.
NoRest Standard
$499/ mo
Flat. The heavy month does not change it. Speed is $899 for twice the rate limits.
Featherless, 32B abliterated
$2,409
Same month on roslein/Qwen3-32B-abliterated. Developer rates, $0.408 in and $1.972 out per million.
abliteration.ai, base
$3,888
abliterated-model at $1 in and $3 out, after the 10% Scale discount. List price is $4,320.
abliteration.ai, large
$8,554
The 1M-context models at $3 in and $5 out, same 10% off. List price is $9,504.
A few requests a minute is enough
Hold the prompt shape fixed and turn the rate down until the other bill matches Standard.
Featherless on that 32B model passes $499 at about 4.1 requests a minute, sustained. abliteration.ai's base model passes it at about 2.6 a minute. Their large model passes it at about 1.2. Speed, at $899, crosses the 32B Featherless bill around 7.5 a minute and the base abliteration.ai bill around 4.6.
From there the month stays a flat fee. A batch that runs long does not open a second invoice.
Same job, different bill
What you pay for
NoRestA month of the model. Tokens are not a line item.$499 Standard, $899 Speed. 60 or 120 requests a minute. 4 or 8 in flight.
abliteration.aiTokens, plus a small plan that is really a credit.Developer $20, Growth $50, Scale $200. Each plan's price is monthly usage credit, then 2.5%, 5%, or 10% off further usage.
FeatherlessTokens, once you are allowed to use the API.Developer starts at $50 of credit. Rates are per model. Credits do not expire, and they do not shrink a bill this size.
The cheap plan
NoRestThere isn't one. Both plans are the full API, unlimited, 24/7.
abliteration.aiYou can start on prepaid credit with no subscription. The meter is the same meter.
FeatherlessChat is $25 and says unlimited tokens. Their pricing page bars that plan from app traffic, background automation, and benchmarking. The API is the Developer plan.
The model
NoRestOne model, the one on the key. Same weights every day.
abliteration.aiThree hosted models. Base is 256K with image input. Large and large v2 are text, 1M context. Large v2 is their GLM 5.3.
FeatherlessA catalogue, 40,000 and up, including a lot of other people's abliterations. You pin a repo. Quality, context, and the rate are whatever that card says today.
When it gets busy
NoRestThe rate limit answers 429, with Retry-After. The month's price does not move.
abliteration.aiYou pay the tokens, up to a daily organisation cap that this workload does not reach (2 billion tokens a day on their $20 tier). Past the cap, you wait until the next day.
FeatherlessShared GPUs. Their own FAQ: a saturated model returns 503, the attempt is free, and you retry or switch models. Popular models saturate first.
Prompts
NoRestNot stored. They live in GPU memory for the request.
abliteration.aiNot stored by default. They keep token counts and error codes for billing.
FeatherlessTheir pricing page says prompts and completions are never stored.
You can buy it today
NoRestNo. Every GPU is allocated. The waitlist is the door.
abliteration.aiYes. Card, key, go.
FeatherlessYes. Card, key, go.
abliteration.ai is the closer product. Hosted abliterated weights, an OpenAI-compatible API, and a policy gateway if you want a second system deciding allow, refuse, rewrite, redact, or escalate. That gateway is a real feature. We don't sell it. They also publish a 1M context on the large models, and they will provision you this afternoon.
Featherless is a warehouse. If you want to swap models every week, it is the better shelf. The $25 plan is a chat subscription with a 32K window and 4 concurrent units, and they will cancel it if you point an app at it. The API price is the card on the model you picked, and a cold or saturated model is your problem to route around.
We are the better buy when the key is infrastructure. One model, on all day, and a bill you can read in September on the day you signed. For the month in the first row, it is the smaller number by a lot.
How the dollars were built, and where the rates came from.
- Calls = requests per minute × 43,200 minutes. Tokens = calls × 2,000 in and × 1,000 out. No cached-input discount, because these calls don't repeat a long prefix.
- abliteration.ai list rates, 10 Oct 2026: abliterated-model $1 / $3 per million in and out. Large and large v2 $3 / $5. Scale takes 10% off. Cash paid equals the discounted usage once that usage is past the $200 credit. abliteration.ai/pricing, docs rates, rate limits.
- Featherless Developer, same day, model card roslein/Qwen3-32B-abliterated: input $0.408, output $1.972 per million. Plans. The $25 Chat plan excludes API traffic.
- NoRest prices are the ones on our pricing section. Standard $499, Speed $899, unlimited tokens, 24/7.