Uncensored models for autonomous AI agents

For an autonomous AI agent, run Kimi K3.1 Pro on long tasks and DeepSeek V4.1 Flash 1017 where cost per step matters. Every model below reads 1,048,576 tokens and supports tool calling.

Updated November 18, 2026By unrestrictedaiagent

Models and run cost

A step is 20,000 input and 1,000 output tokens at list prices, read November 18, 2026.

Per 1M in / out$2.00 / $4.00

GLM-5.3

Z.ai, 753B

Input
Text
Max output
131,072
100 steps
$4.40
100 steps, 80% cached
$1.68
Per 1M in / out$0.24 / $1.00

GLM-5.3 Flash

Z.ai, 320B total, 18B active

Input
Text, images
Max output
131,072
100 steps
$0.58
100 steps, 80% cached
$0.29

Picks by job

Multi-hour runs

Kimi K3.1 Pro

943,718-token max output and 2,780B parameters.

Browser agents

Kimi K3.1 Pro

Takes text, images and video.

High-volume workflows

DeepSeek V4.1 Flash 1017

$0.24 in, $0.038 cached, $0.90 out.

Text-only tool chains

GLM-5.3 Flash

$0.24 input, 320B total, 18B active.

Run your own

Open-weight abliterated models from 8B to 36B parameters run on one GPU.

Running agents locally