The Inference Directory
Explore directory

Abliteration.ai

Verified

Abliteration.ai is listed as a dedicated lab provider with 4 models in the cybersec directory.

Dedicated lab No KYC OpenAI + Anthropic
Visit provider Compare scores

At a glance

Record
Operator
Abliteration (US)
Group
Dedicated lab
KYC
No KYC
API shape
OpenAI + Anthropic
Founded
2026

Summary

Scope

Card only. Most expensive per token in the category. Own FAQ says some prompts still refuse.

No prompt/output retention by default; cached input billed at 10%.

Source: abliteration.ai/pricing ↗ · KYC scan 4 November 2026

API access

OpenAI + Anthropic

Payments

Card

KYC

No KYC

Best score

CyberBench 72.1

Key models

4 listed
Abliterated Large v2 (GLM-5.3)Frontier 1M · Text only; reasoning cannot be disabled, only hidden. Abliterated $3.00 / $5.00
Abliterated Large Flash (DeepSeek V4.1 Flash)Frontier 1M · Text + image input. Abliterated $1.00 / $2.00
Abliterated Model (multimodal) 256K · Text + image input. Abliterated $1.00 / $3.00
Abliterated Large (deprecated) 1M · Kept for existing integrations. Abliterated $3.00 / $5.00

Pricing

List rates · USD per 1M tokens
Model Type Input price Caching price Output price
Abliterated Large v2 (GLM-5.3)1M · Text only; reasoning cannot be disabled, only hidden. Frontier $3.00 $0.3 $5.00
Abliterated Large Flash (DeepSeek V4.1 Flash)1M · Text + image input. Frontier $1.00 $0.1 $2.00
Abliterated Model (multimodal)256K · Text + image input. Model $1.00 $0.1 $3.00
Abliterated Large (deprecated)1M · Kept for existing integrations. Model $3.00 $0.3 $5.00

List rates. Cached input shown where published; "-" means the provider does not publish a cache rate. Workload comparison uses 1M input plus 250K output. Per 1M tokens

Access

Rails & limits
CardSupported
CryptoNot supported
RailsCard
Minimum top-upNot published

Most expensive per token in the category; own FAQ admits some prompts still refuse.

KYC

No KYC

No KYC published; Stripe card only.

No ID step at signup or checkout; email + card/crypto only

Vendor source ↗ · Basis: Directory scan

Scores

Published cyber benchmarks
ModelCyberBenchCyberOffenseCyberGym
Abliterated Large v2 (GLM-5.3) 72.1 - 84.5
Abliterated Large Flash (DeepSeek V4.1 Flash) 73.7 72.8 88.1

Performance

Coming soon
Latency

Time to first token

Median and p95 response times for short and long prompts, measured from repeatable probes rather than vendor claims.

Coming soon
Throughput

Tokens per second

Streaming output rate and request throughput under steady load, with concurrency and context length recorded.

Coming soon
Efficiency

Cost per workload

Blended cost for a fixed workload, including cache reads and output tokens, so speed and price can be compared together.

Coming soon

Independent performance probes are being added. Vendor-reported numbers are not mixed into this section.