CoreThink is a state-of-the-art inference layer built on a novel inference method called General Symbolics. It differs from existing inference paradigms such as test-time scaling, supervised fine-tuning (SFT), and reinforcement learning with verifiable rewards (RLVR). The CoreThink General Symbolic Reasoner (GSR) is structured around three key use cases: tool invocation, code generation, and planning, and demonstrates outstanding performance across seven benchmarks in each domain. Specifically, it achieved state-of-the-art performance (SOTA) scores of 66.66% on Livecodebench v6, 89% on Instruction-Following Evals, and 24.4% on ARC-AGI-2. Furthermore, we present an agent coding IDE developed using the principles of General Symbolics, achieving a state-of-the-art accuracy of 62.3% on SWE-Bench Lite. This performance improvement was achieved without any fine-tuning or training costs. The CoreThink inference layer is designed to deliver pure performance gains, ensuring that the accuracy of the model's inference tasks never degrades. The authors argue that existing methods will ultimately lead to diminishing returns in LLM performance, necessitating the development of new inference techniques. This technical report details the CoreThink approach at a high level and the availability of CoreThink models for inference-intensive use cases.
