Meta published LLaMA (7B-65B parameters) in February 2023, proving that smaller models trained on public data alone could match GPT-3 (175B) on most benchmarks. The weights leaked within days and spawned Alpaca, Vicuna, and dozens of community variants, lowering the technical and capital barriers to LLM development and deployment. This event established open-weight models as a durable competitive alternative to closed-source APIs, directly pressuring proprietary vendors on pricing and market access. For financial institutions, this shift implies that frontier-model economics may concentrate differently: API-layer defensibility weakens where open alternatives reach task-parity, but infrastructure, deployment-efficiency tooling, and domain fine-tuning become durable margins.
239 other tracked developments touching the same claims.