Vietnam and Project Tapestry have affirmed an intention to collaborate on a sovereign AI model covering Vietnamese language, culture, knowledge and national use cases. The AI Alliance announced the discussions with Vietnamese leaders on September 30 through PRNewswire. Vietnam was already an early participant. The release does not establish a signed development contract or a deployed national model.
The project completed its initial technical milestone on September 1 and published findings on September 10. Separate distributed training proofs operated in India, Australia and AWS regions, with no Vietnamese training site identified. The larger run continued training OLMo 3 7B on 12 billion tokens across Sydney and Virginia, using 8 H100 GPUs at each site and aggregation in Ohio. Partitioned training data stayed at the training sites while model parameters were exchanged.
A separate study tuned a 3 billion parameter Llama 3.2 Instruct model using LoRA. Its distance to Vietnam on the Inglehart Welzel cultural map fell from 2.46 to 1.35, approximately 45% closer. Full MMLU scores moved from 63.2% to 62.4%, a decline the preliminary report described as not statistically significant. This was a feasibility study of a model and a target culture, not evidence of nationwide performance.
An independent validation harness using Arabic corpora found stronger shifts in survey answers than in open ended behaviour. That experiment does not directly validate the Vietnam result, but shows why a cultural survey score needs behavioural testing. Data governance and management requirements reached version 0.1, with further work assigned to dedicated groups.
The next milestone runs from September through November 2026, extending consortium training, cultural alignment and data governance. The roadmap targets a first base model by the end of 2026 and early deployment in the first half of 2027. Progress depends on technical proof, partner commitments, compute, data readiness and funding. The announcement discloses no Vietnam specific compute allocation or funding commitment.