Models

Abliterated models, plus the official DeepSeek API

Full-weight abliterated builds of frontier open-weight models, served on dedicated GPUs. Text only, 256K context, reasoning effort set per request. Compare prices.

FP8 · Zero additional quantization256K context64K max outputText only

GLM-5.3 Flash Abliterated

glm-5.3-flash-abliterated

Agentic all-rounder with strong terminal and tool-use scores.

Input: $0.0977
Output: $0.3421
Cached input: $0.0098

USD per 1M tokens

Speed — · measured, last 24 h

Benchmarks: base model, vendor-reported unless noted (zai-org/GLM-5.3-Flash)

Terminal-Bench 2.1: 84.3source
DeepSWE v1.1: 63.4source
HLE (with tools): 55.3source
GPQA-Diamond: 92.17NVIDIA BF16 run

Abliteration effect: MMLU (n=300) 83.3 → 82.7 (orcarouter build, reported by its publisher, source).

Model details

Official DeepSeek API

DeepSeek's own models, run by DeepSeek

Same API key, same balance and the same endpoints, served through DeepSeek's official API instead of our dedicated GPUs.

These are DeepSeek's stock models, served through DeepSeek's official API. They are not abliterated and keep DeepSeek's own refusals.

DeepSeek may also decline some requests on its side under its own terms of use. See the DeepSeek Terms of Use.

Prompts and outputs for these models are sent to DeepSeek, processed and stored in the People's Republic of China under DeepSeek's privacy policy, and may be used by DeepSeek to train its models. Our no-storage promise covers our own systems only. See the DeepSeek Privacy Policy and our privacy policy.

Official DeepSeek API1M context384K max outputText only

DeepSeek V4.1 Flash

deepseek-v4.1-flash

DeepSeek's own stock Flash model, run by DeepSeek through its official API.

Stock model, not abliterated: it keeps DeepSeek's own refusals, and DeepSeek stores prompts in China and may train on them. Details

Input: $0.30
Output: $1.20
Cached input: $0.006

USD per 1M tokens

Speed — · measured, last 24 h

Model details

Official DeepSeek API1M context384K max outputText only

DeepSeek V4 Pro

deepseek-v4-pro

DeepSeek's own stock Pro model, run by DeepSeek through its official API.

Stock model, not abliterated: it keeps DeepSeek's own refusals, and DeepSeek stores prompts in China and may train on them. Details

Input: $1.32
Output: $3.96
Cached input: $0.044

USD per 1M tokens

Speed — · measured, last 24 h

Model details

Machine-readable list: /models.json or GET https://api.abliteratedllm.dev/v1/models.