SINGAPORE, Feb. 11, 2026 (GLOBE NEWSWIRE) —
- Up to 80% lower training costs compared to traditional cloud-based reinforcement learning.
- State-of-the-art performance achieved without relying exclusively on centralized data centers.
- More experiments per dollar, shifting AI progress from infrastructure limits to research speed.
Gradient, an AI infrastrucUp to 80% lower training costs compared to traditional cloud-based reinforcement ture provider, today announced the release of Echo-2, a new reinforcement learning system designed to dramatically reduce the cost and infrastructure required to train advanced AI models. The launch targets one of the biggest emerging bottlenecks in AI development, where progress is no longer limited by ideas or talent, but by access to expensive, centralized computing power.
The first wave of generative AI was driven by training models on massive datasets. The next leap, however, depends on post-training, where models improve through repeated trial and error. This process, known as reinforcement learning, helps AI systems reason, plan, and adapt. It’s also one of the priciest stages of AI development. It often needs big, energy-hungry data centres, making advanced training too costly for most organisations.
Echo-2 is designed to change that dynamic. Rather than forcing all training to run inside tightly controlled clusters, the system allows reinforcement learning workloads to be spread across a wide range of hardware. In early benchmarks, Gradient reports up to 80% cost reductions compared to traditional cloud-based approaches, while matching or exceeding performance on reasoning and agent-based tasks. The result means that teams can run far more experiments, learn faster, and improve models without relying exclusively on hyperscale infrastructure.
“AI progress is no longer limited by ambition, but by infrastructure,” said Eric Yang, Co-Founder and CEO of Gradient. “Reinforcement learning is becoming the engine of real intelligence, yet today it is locked behind enormous data center costs. Echo-2 lowers the cost of experimentation, so more teams can build, test, and improve AI systems without needing access to hyperscale cloud infrastructure.”
The release comes as governments and enterprises face growing constraints on power availability, environmental impact, and data control associated with large AI data centers. As AI becomes more central to how organizations operate and compete, the ability to train and improve models without concentrating massive workloads in a single location is becoming increasingly important. Systems like Echo-2 offer an alternative path that supports continued AI progress while easing pressure on centralized infrastructure.
Echo-2 builds on Gradient’s broader work in distributed AI infrastructure, following the company’s earlier release of Parallax, which enables large AI models to run across multiple machines, supported by its Lattica networking layer that handles data communication between distributed systems. Together, these systems aim to reduce dependence on centralized compute by allowing AI models to be trained, deployed, and improved using existing hardware across distributed environments.
By lowering costs and increasing flexibility, Echo-2 allows research teams to move faster and run more experiments, while giving enterprises a way to reduce long-term reliance on expensive cloud commitments. As cost, power, and infrastructure constraints increasingly shape the future of AI, Gradient says Echo-2 broadens access to advanced reinforcement learning at a critical moment. The company is productising Echo-2 as part of a broader distributed reinforcement learning platform, alongside the launch of Logits, an RL-as-a-Service platform built on its distributed architecture, with enterprise access expected later in 2026.
– ENDS –
About Gradient:
Gradient is an AI R&D lab dedicated to building open intelligence through a fully decentralized infrastructure – OIS (Open Intelligence Stack), encompassing distributed training, serving, agentic systems, and more.
Backed by top investors and a team of world-class researchers, Gradient is committed to releasing more frontier research that will unlock a future where intelligence can be assembled, scaled, and evolved by anyone, anywhere.
