INT21 Try AI-Built Engine

INT21 / Blog

News and insights.

Company updates and technical perspectives on self-improving compute infrastructure. For media and analyst queries, contact press@int21.ai.

Training infrastructure progress: 11.5× faster than eager PyTorch FSDP2 and 21.5% faster than SwarmOS-tuned PyTorch FSDP2.

Research /

SwarmOS and GPT-6 Astra Built a Qwen3.8 Trainer 11x Faster Than PyTorch

SwarmOS and GPT-6 Astra generated a Rust/CUDA Qwen3.8-27B trainer with 11.5× the throughput of eager PyTorch FSDP2 and 21.5% higher throughput than the tuned baseline.

Read article
Grouped bar chart showing GPT-5.6-Sol improving from 13.3% to 100% and GPT-5.6-Luna improving from 0% to 56% with SwarmOS, set against faint ARC-style pixel-puzzle motifs.

Research /

Pushing GPT-5.6-Sol from 13.3% to 100% on ARC-AGI-3 Public

SwarmOS pushed GPT-5.6-Sol from a 13.3% baseline to 100% RHAE on ARC-AGI-3 Public, showing how orchestration can multiply long-horizon agent capability.

Read article
Three lower-is-better benchmark lanes comparing INT21 with SGLang and vLLM: 10.80 versus 12.03 seconds for MiniMax H3 video, 4.24 versus 4.91 seconds for MiniMax Music 3, and 0.252 versus 0.301 and 0.326 real-time factor for Fish Audio S2 Pro speech.

Engineering /

Addressing the Inference Bottleneck

INT21's agent swarms built live inference engines for video, music, and speech generation—without hand-built frameworks or human expert tuning.

Read article
A dark GPU compute substrate where cool candidate paths pass through verification gates and return as amber validated improvements.

Perspective /

What Is Self-Improving Infrastructure? Defining a New Category

Self-Improving Infrastructure uses AI agent swarms to continuously improve the software AI runs on, making it faster, cheaper, and more reliable.

Read article
A central orchestration core coordinating a recursive network of specialized AI agents.

Research /

What Muse Spark 1.1 Taught Us About Enterprise Agent Architecture

We ran Meta's Muse Spark 1.1 inside SwarmOS. The result shows why the durable enterprise advantage is the orchestration layer, not any single model.

Read article
A large vertical context window dissolving into a coordinated network of specialized AI agents across compute infrastructure.

Engineering /

Stop Waiting for a Bigger Context Window

Bigger context windows don't solve long context; multi-agent orchestration does. Here's how we tackle workloads involving tens of millions of tokens.

Read article
A cinematic view of interconnected compute infrastructure.

Launch /

Introducing INT21 and PTX Kernel Factory

We are building self-improving AI systems for the software beneath modern AI. Our first product generates and optimizes low-level GPU software, then proves its work with tests and benchmarks.

Read article