SWE-2 vs. Composer 2.5

SWE-2 is Cognition’s coding model for software engineering agents, developed to improve the balance between capability, reasoning cost, and execution efficiency. The model is post-trained from Kimi K3, a multi-trillion-parameter model that had already received extensive reinforcement learning for agentic coding. Cognition further trained SWE-2 with a reinforcement learning algorithm that optimizes several reasoning-effort levels during a single training run. These effort levels let users trade off speed and cost against deeper planning, codebase exploration, and verification for more difficult assignments. SWE-2 is designed to reduce the over-exploration seen in earlier models by identifying relevant files and implementation paths more quickly. Its software engineering abilities include repository analysis, code writing and editing, debugging, testing, build and lint workflows, terminal tasks, and verification of completed work. The model places additional emphasis on writing end-to-end tests, catching edge cases and regressions, and gathering evidence instead of simply accepting assumptions in a prompt. Cognition’s training approach also uses cost penalties tied to the model’s performance frontier, length-weighted reward baselines, speculative decoding improvements, low-precision inference techniques, and expanded reinforcement learning data. Training data includes more diverse repositories, additional instruction-following requirements, and iterative verifier improvements designed to reduce reward hacking and false validation. SWE-2 is benchmarked against models such as GPT-6 Astra, GPT-5.6 Sol, Fable 5.1, Grok 4.6, and Kimi K3, with Cognition positioning it around strong coding performance at substantially lower cost. SWE-2 is intended for use across Cognition’s Devin ecosystem, including Desktop and CLI, with rollout to Devin Web and Fusion.

Composer 2.5 is Cursor’s newest AI-powered coding model, designed to significantly improve software development productivity through stronger reasoning, enhanced collaboration, and better handling of complex engineering tasks. Compared to Composer 2, the new release delivers major gains in sustained coding performance, allowing developers to work on larger and more complicated projects with improved reliability. The model was trained using expanded compute resources, more advanced reinforcement learning environments, and additional optimization techniques focused on both intelligence and usability. Cursor also refined behavioral aspects of the AI, including communication style and effort calibration, to make interactions feel more natural and productive during real-world coding sessions. A major feature of Composer 2.5 is its targeted reinforcement learning system with textual feedback, which provides localized corrections during training when the model makes mistakes such as invalid tool calls or style violations. This approach helps the AI understand exactly where errors occur and improves its decision-making more effectively than broad reward signals alone. The company further strengthened the model by training it on 25 times more synthetic coding tasks than Composer 2, exposing it to a wider range of difficult engineering challenges and edge cases. These synthetic tasks included feature deletion exercises where the model had to reconstruct missing functionality in real codebases using automated tests as validation signals. During large-scale training, Composer 2.5 demonstrated advanced problem-solving capabilities by reverse-engineering cached data and decompiling Java bytecode to recover deleted APIs in synthetic environments. Cursor also implemented sophisticated distributed training systems such as Sharded Muon and dual mesh HSDP, allowing efficient optimization across extremely large AI models and infrastructure clusters.