IOV LABSIOV LABS, an AI research lab, launched in Seoul on May 1, 2026, with a stated focus on building open-source developer tools and publishing reproducible benchmarks. The lab said it would release its work publicly in both English and Korean, and that its first project is 0x-lang, an open-source programming language aimed at AI code generation. The name IOV stands for Input, Output, Value, which the lab describes as the path it cares about most: an input becomes an output, and an output becomes value only once it is something a person actually relies on.
Why now
The lab said its work is aimed at the gap between AI demonstrations and tools that people use in daily work. Modern language models are capable, it argued, but the surrounding infrastructure for making their output reliable, inexpensive and verifiable is still largely missing, and the representations those models read and write are mostly accidents of training data rather than deliberate designs. The lab also pointed to a recurring pattern in the field, in which benchmarks are chosen to impress rather than to inform, and said it intends to work against that pattern by measuring its own results with numbers anyone can reproduce.
What it makes
According to the lab, its output falls into three categories. The first is open-source tools meant to be usable within a few minutes, on the view that a tool which feels useful immediately is worth more than a powerful one nobody finishes setting up. The second is small public demonstrations that make an idea legible at a glance and are worth sharing. The third is benchmarks and studies in which every claim is backed by a reproducible number, and in which negative results, including approaches that did not work, are reported alongside the successes. The lab said an honest record is more useful to other developers than a selective one.
First project: 0x-lang
The lab's first project, 0x-lang, is an open-source programming language that compiles a single source file into React, Vue 3, Svelte 5, React Native, Express and Terraform, and ships with a language server and a Model Context Protocol server. IOV LABS has published a token benchmark and a code-generation study for it, reporting that 0x source uses roughly 2.4 times fewer tokens than the React it produces, and that with constrained decoding and three compiler fixes a model compiled valid 0x on five of five tasks on the first attempt, holding at seven of eight on an unseen set. The lab said the benchmark is reproducible from the repository with a single command.
How it works
IOV LABS said its research is funded by the founder's own resources rather than outside investment, and that abundant, low-cost compute lets it run experiments freely. Much of its work, the lab said, begins as exploratory research with no obvious connection to a product, some of which it expects to develop into standalone tools over time. The lab added that it operates independently and does not draw on materials from the founder's other commercial work, and that everything it ships is open and bilingual from the first commit, so that neither an English nor a Korean audience receives a thinner version of the story.
Honesty is the moat: reproducible benchmarks, and a frank account of what did not work.
The founder
The lab was founded by Han Kim, who serves as chief executive. "Most of the industry is fascinated by the model that turns input into output," he said. "We are more interested in the last step, where an output quietly becomes something a person depends on, because that is where technology either earns its place or does not." He said the lab would measure that progress in public, and invited developers to judge it by the work rather than the announcement. IOV LABS noted that its name refers to Input, Output, Value and that it is unrelated to IOV Labs, the blockchain company associated with RSK.
What is next
In the near term, the lab said it would continue developing 0x-lang and expand a collection of smaller tools and MCP servers it calls Constellation, releasing updates every week or two. Each release will be documented as it happens, with notes published in both languages. The lab invited developers to follow the projects on GitHub, reproduce its benchmarks, file issues, and contribute.