Deep Dive
1. Major zkML Milestones & Optimizations (September 2025)
Overview: This update made Lagrange's DeepProve system compatible with the latest, more efficient AI models like Google's Gemma3. It also introduced backend optimizations that make proof generation faster and cheaper.
The team successfully proved inference for the 270M-parameter Gemma3 model, a significant technical milestone. To achieve this, they extended DeepProve's framework to handle new AI architecture features. Key optimizations were also deployed: a new in-house graph architecture improves reliability for distributed proving, a unified "Einsum" layer simplifies and accelerates linear operations, and a tensor deduplication feature eliminates redundant work, cutting proving time and memory use.
What this means: This is bullish for $LA because it demonstrates the project's technical leadership in verifiable AI. The upgrades mean the network can handle more complex AI verification tasks faster and at a lower cost, which could drive greater demand for proof generation services that use the LA token.
(Source)
Overview: This refactor dramatically improved DeepProve's proving speed and efficiency, enabling it to verify much longer AI inferences and setting it up to run on various hardware, from personal computers to large server clusters.
The update enabled full-sequence (1024 token) proofs for GPT-2 on the same hardware previously used for much shorter runs, showcasing a 25x throughput improvement. The core system was rebuilt on a newer, more efficient cryptographic base (Scroll's "Ceno"), which doubled proving speed and slashed memory use by 10x. A new memory management framework was also introduced, making the prover portable across devices, and work began to migrate AI inference calculations from CPU to GPU for even greater speed.
What this means: This is bullish for $LA as it directly enhances the network's utility and scalability. Faster, cheaper proofs make the service more attractive to developers, while hardware flexibility allows the network to grow and meet rising demand, potentially increasing token usage.
(Source)
Conclusion
Lagrange's recent codebase updates reveal a clear focus on scaling its core zkML technology for practical, high-demand use cases in verifiable AI. With proven compatibility for cutting-edge models and major performance gains, the foundation is strengthening for increased network utility. How will these technical advancements translate into measurable on-chain demand for $LA in the coming quarters?