GSMA today unveiled Open Telco AI, a global initiative designed to drive telco-grade AI development through an open portal offering models, open datasets, compute toolchains and evaluation benchmarks. The portal centralizes resources for operators, vendors, developers and academics to collaborate on telecom-specific AI, with progress measured by a multi-task Telco Capability Index.

As a founding supporter, AT&T is releasing a family of open telecommunication models trained on publicly available data to ensure hardware and cloud agnosticism. These models span multiple sizes and architectures, targeting tasks from network troubleshooting to standards interpretation and aiming to boost accuracy where only 16% of current GenAI deployments address network operations.

AMD is committing GPU platforms for model training, fine-tuning, inference and evaluation, while cloud partner TensorWave provides additional compute capacity and open toolchains. This combined infrastructure enables projects of varying scale to access enterprise-grade AI compute from core to edge environments.

The initiative unites operators, vendors and academic institutions through community programs like the AI Telco Troubleshooting Challenge, which drew over 1,000 participants. Founding partners will contribute models, datasets and use cases, and submit results to a live leaderboard that evaluates performance across seven telecom-specific benchmarks.