The barrier to entry for autonomous, high-level research tools is collapsing. The NLP team at Ohio State University has unveiled QUEST-35B, an open-source Deep Research agent designed to rival frontier closed systems, proving that state-of-the-art reasoning can be achieved without massive corporate budgets.
Academic Budget, Frontier Performance
What makes QUEST-35B significant is the total transparency of its origin. The researchers have open-sourced everything: weights, training code, and the datasets used. The model was trained using approximately 32 H100 GPUs and 8,000 synthetic samples. The released recipe provides a comprehensive roadmap covering SFT (Supervised Fine-Tuning), mid-training, and RL (Reinforcement Learning) phases.
Scalability for Developers
The project offers a versatile range of model sizes, from 2B to 35B parameters, allowing developers to deploy deep research capabilities across various hardware constraints. This move aligns with a broader shift toward open agentic AI, as seen with Harness-1, which has shown competitive results in complex information retrieval benchmarks.
The Shift Toward Autonomous Open AI
QUEST enters a landscape where open-source models are no longer just smaller versions of proprietary ones, but specialized tools. From Tencent's SearchAgent-8B to the disruptive efficiency of DeepSeek, the industry is moving toward a future where high-reasoning agents can be self-hosted. This shift reduces reliance on expensive API ecosystems and empowers researchers to build customized, private, and scalable autonomous agents.
No comments yet. Be the first!