Introducing Kili 2.5
Our flagship 230B-parameter model with a 590K-token context window — delivering frontier reasoning, native tool use, and multimodal comprehension for the most demanding workloads.
- Date
- June 9, 2026
- Category
- Announcement
The language models your hardest problems deserve
Merkium is an AI research laboratory. We develop the Kili family of large language models — purpose-built for safety, steerability, and real-world utility at scale.
Released June 9, 2026
Kili 2.5 — 230B parameters, 590K context
Learn moreLatest releases
Kili 2.1 — stability release
A targeted refinement improving instruction adherence, structured output, and tool-use reliability across extended agentic sessions.
- Date
- May 31, 2026
- Category
- Model update
The Merkium Alignment Index
A systematic benchmark measuring how consistently frontier models preserve human intent across high-stakes, multi-step domains.
- Date
- May 1, 2026
- Category
- Research