Laguna M.1
A public foundation model from PoolsideAI for agentic coding, mainly helping developers and researchers build code agents, long-context coding experiments, and self-hosted or hosted inference setups.
Tool overview
Adoption-wise, Laguna M.1 looks worth tracking and trying if you care about open-weight coding models, but the current evidence supports availability and positioning more strongly than broad independent proof of day-to-day performance. It is not an AI IDE, and not a turnkey autonomous software engineer. A better analogy is an open-weight coding-agent foundation model plus Poolside’s surrounding agent harness.
In practice, it seems useful for two things: plugging into existing coding workflows through Poolside API, OpenRouter, or tools like Cline for long-context and agent-style coding tasks; and experimenting with the released weights in self-hosted setups. Official and near-official sources support 256K context, released base and post-trained checkpoints, Apache 2.0 licensing, and a technical report. There is also a social demo of a 3-bit MLX build on Apple Silicon at about 26 tok/s with roughly 100GB peak memory, but that is a community-style demo, not a stable hardware promise.
On cost and difficulty, the official evidence confirms public weights and access through Poolside API and OpenRouter, but not a durable pricing commitment in these sources.