September 3, 2026, (Inside AI) — A research institute in Abu Dhabi has published six artificial intelligence models alongside the training data, code, and methods used to build them. The release from IFM, announced Thursday, includes model weights, training data, code, methodologies, and intermediate checkpoints.
Founder Eric Xing said the package lets researchers retrace the models’ development and reproduce results. The K2 Horizon family spans from a model for smartwatches to a 375-billion-parameter system for enterprise use.
This move challenges a growing industry trend toward secrecy. Many leading developers release only model weights or keep everything private. IFM’s release goes beyond the open-weight approach used by some Chinese developers. It also contrasts with the closed practices of OpenAI and Anthropic, which do not release models or disclose training details.
Why Full Openness Remains Rare
Open-weight releases have become common, but full training transparency is not. Most open models share weights and some code, yet withhold datasets or training recipes. This limits independent verification and reproduction.
Read: Gnani Unveils Sovereign AI Stack Artha With Open-Weight Model and Enterprise Agents
IFM’s package includes intermediate checkpoints, a rare addition. These snapshots let researchers study how models evolve during training. Such detail supports audits, safety research, and reproducibility.
Xing framed the release as a reference point. “Our goal with this release is to establish a reference point for what a truly open model release can look like,” Xing said.
He also addressed regulators and advocates. “Openness and competitive performance are not mutually exclusive,” he said.
UAE’s AI Ambitions Get a Boost
The launch is part of the UAE’s push to become a global AI hub. Abu Dhabi has invested heavily in compute, talent, and research infrastructure. Fully open releases could attract researchers who value transparency.
The model family targets diverse hardware. One model is designed for smartwatches and constrained devices. The largest has 375 billion parameters for enterprise deployments. This range shows an effort to serve both edge and cloud use cases.
Inside AI could not independently verify the performance claims. No benchmark scores were provided in the announcement. The release’s impact will depend on whether the community adopts and tests these models.
Full openness carries risks. Training data may include copyrighted or personal information. Releasing code and checkpoints can enable misuse. IFM has not detailed its data filtering or safety evaluations.
Read: OpenAI Gives 100,000 Researchers Free Access to GPT-5.6 Models
Still, the move adds pressure on other developers. As governments debate AI transparency rules, fully open releases provide a concrete example. They may influence future policy discussions and industry norms.
