SWE-2 vs. GPT‑6 Cyber

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.

GPT-6 Cyber is a frontier AI cybersecurity model developed by OpenAI to support autonomous vulnerability research, exploit validation, secure software development, and advanced security testing. The model is specifically trained to investigate complex software environments, discover exploitable vulnerabilities, and help security professionals develop effective remediation strategies. Its capabilities include source code analysis, reverse engineering, vulnerability reproduction, exploit development, and investigation of security weaknesses across large software projects. GPT-6 Cyber can conduct extended security research workflows that involve identifying potential attack paths, testing hypotheses, validating technical findings, and producing reproducible evidence for security teams. During OpenAI's research campaigns, the model identified 81 previously unknown high-severity Firefox vulnerabilities and helped develop candidate patches for affected components. It also reproduced 1,724 Linux kernel vulnerabilities and generated working exploits for 948 of them, demonstrating its ability to validate security issues at scale. On the ExploitGym benchmark, GPT-6 Cyber achieved a 48.5% success rate compared with 32.1% for GPT-5.6 Cyber, while achieving 40.3% vulnerability recall in an internal web penetration testing evaluation. The model incorporates specialized alignment training to follow authorized testing scopes, respect security restrictions, and avoid exploiting systems that fall outside the researcher's stated objectives. In OpenAI's ExploitGym Honeypot evaluation, GPT-6 Cyber avoided attacking out-of-scope services in 99.93% of tests without deployment safety instructions and 100% of tests when those instructions were included. Its deployment safeguards include misalignment monitoring, misuse detection, IP allowlisting for API access, human oversight recommendations, and additional protections for operational technology and critical infrastructure systems.