DeepSeek V3.2 Claims GPT‑5‑Level Power With 90% Lower Training Cost

AI outlet Tech Wire Asia reports that DeepSeek’s new V3.2 model is benchmarking at or near GPT‑5‑class performance on key tasks while using roughly 90% less training compute than rival frontier systems. The company credits architectural tweaks, data curation and heavy optimization for the dramatic efficiency gains, framing the model as proof that smarter training can beat brute-force scale. This positions DeepSeek as a serious competitor in the race to deliver powerful, cost‑effective foundation models.​

Lower training costs matter because they can translate into cheaper inference, more frequent upgrades and the ability for smaller players or regions to host strong models on their own infrastructure. Analysts note that if these claims hold up under independent evaluation, major cloud and enterprise buyers may explore DeepSeek as an alternative or complement to US and European providers. The model’s performance on multilingual, coding and reasoning benchmarks will be critical in that decision.​

For teams building AI products, this development underscores the importance of comparing not just raw quality, but also cost per token, latency and deployment flexibility. Before committing to a single model family, consider pilot tests across DeepSeek and other leading options on your real workloads. Sign up for technical reports, API waitlists and benchmark dashboards to track how V3.2 evolves.​