Blog
How Much Does the Agent Harness Matter?
We ran the same DeepSeek model slug on ten Terminal-Bench 2.1 tasks with five agent harnesses. The harness changed pass rate, cost, speed, and memory use.
Claude Code Cut Their System Prompt by 80%. Does That Work for Small Models Too?
We A/B tested Ante's half-size system prompt on deepseek-v4-flash across the full terminal-bench 2.1 suite: no measurable performance change, and among the 69 tasks whose outcome stayed the same, the short-prompt run's median input-token count was 32% lower.
From Arcade to Living Room: Offline Coding Models Hit Their Console Moment
On Terminal-Bench 2.0, open-weight 27B–35B models now match what the hosted frontier was posting 6–8 months ago — close enough that, for privacy-first and air-gapped teams, running it locally is finally a serious engineering choice.
Introduce Ante: Self-Contained Agent That Self-Organize
Today Anthropic "open-sourced" Claude Code — and it's the perfect day to introduce Ante, the precursor of our mission, a self-contained agent built from first principles.
How to Achieve #1 on Terminal Bench
and Why We Can't Have Nice Things: A forensic analysis of benchmark manipulation on Terminal Bench 2.
Abliteration: Declaration of Independence from Excessive Model Restriction
A Guide to Relaxing Moderation in Open-Source Language Models.
Neural Cellular Automata (NCA) - Interactive Demo
An interactive demonstration of Neural Cellular Automata with real-time pattern generation and user interaction.
Sovereign Compute and Network State
Second Amendment of the AI Era
The Crypto Way
Bold and boundless
Antigma Manifesto
To openness and clarity