FLock.io’s open-source decision model, built on Qwen, outperforms larger LLMs Built on Qwen3.5-2B and with Qwen and Alibaba Cloud’s support, FLock.io’s new, fast, lightweight System One/decision model delivers sub-31ms local inference, achieving up to 5x the performance of enormous closed-source models on decision-making tasks. The news follows a wave of new decision models such as Typesafe's Jev and Microsoft-Decision-1. FLock.io has unveiled an open-source small decision model for fast, low-cost decision-making, instead of writing text. THIS/THAT Model, built on Qwen3.5-2B and trained on synthetic data, has outperformed frontier larger LLMs, despite being an estimated <0.04% the size. The launch deepens the partnership between Qwen and Alibaba Cloud and FLock.io, a leading privacy-preserving AI solution provider. This specialised model is designed for tasks where quick decisions and data classification matter most – in such cases, the slow, costly natural language generation of chatbots is unnecessary. Three weeks after going live, it has processed 2.5 million decisions. The rise of System One/decision models A new wave of System One models, also known as decision models, such as TypeSafe’s Jev has been disrupting enterprise AI since September. Unlike standard LLMs, which generate open-ended natural language prose, they are for queries that have a predefined set of possible answers such as yes/no. The name comes from the psychological framework of “system 1” thinking (fast, automatic) versus “system 2” thinking (slow, effortful). Chatbots such as ChatGPT do the latter. Outperforms frontier larger LLMs at the edge In FLock.io’s tests, version 1.2 of the new specialised model significantly outperformed GPT-5.6 and Claude Opus on several tasks. It achieved a 87.8% success rate across 1,710 complex decision-making tasks, up from 40.6% in benchmarking. In budget-constrained optimisation, its success was 4x that of GPT-5.6 and 5x Claude Opus 5. This was a rapid improvement from 1.0’s tests just a day before, achieved through optimising training data rather than increasing model size. Jiahao Sun, CEO of FLock.io, said: “For many critical enterprise tasks, generating open-ended text is an unnecessary cost and performance bottleneck. With THIS/THAT Model, we are demonstrating that specialized 2B-parameter models running locally can outperform massive cloud LLMs in speed, accuracy, and cost-efficiency – all while keeping data strictly on-device.” THIS/THAT Model’s specs THIS/THAT Model is small enough to run on standard developer hardware and edge devices, meaning data can stay local and latency can be ultra low. It was built on Qwen3.5-2B: a tiny, edge-friendly foundation model that can be run locally on laptops and smartphones. It is lightweight, with only 1.88B parameters (less than an estimated 0.04% the size of Claude Opus 5) and zero generated output tokens. The local inference latency is ultra low, at approximately 31.4 milliseconds, compared to standard LLMs taking seconds. It was trained on Alibaba Cloud using synthetic training data generated using Qwen Max. It drew on key findings from FLock.io’s recent research papers, GraphSynth and CircuitSynth, which were accepted into the prestigious ACL 2026 conference. The takeaways: high efficiency paired with high performance This launch has demonstrated that specialised models can outperform much larger LLMs on specific decision-making tasks while maintaining high inference efficiency. By combining Qwen’s foundation models, Alibaba Cloud’s infrastructure, and FLock.io’s research, FLock.io is developing high-performance, specialised AI models. Background on Qwen and FLock.io’s partnership Qwen, a tier 1 open-source LLM developed by Alibaba Cloud, partnered with FLock.io in June 2025. It marked the first time a major Web2 centralised LLM has chosen to deeply integrate with a decentralised training platform. The partnership leverages federated learning and distributed ledger technology, enabling organisations to fine-tune Qwen models without exposing raw data to a central server. The two companies have co-hosted joint developer initiatives such as the Qwen SKYST Hackathon in 2025 with top Korean universities. About FLock.io FLock.io is an AI research and infrastructure company pioneering enterprise-grade federated learning and distributed AI solutions. Its decentralised federated learning architecture and production-ready platforms (AI Arena, FL Alliance, FOMO and the API Platform) enable organisations to train and deploy their own custom AI models on local hardware while maintaining full data privacy, model ownership, and regulatory alignment by design.