Introducing the Recommended Agent in AI Chat with Codex as the current choice - The JetBrains Blog

JetBrains AI has introduced Codex as the recommended agent for coding tasks within its IDEs, following a comprehensive evaluation of multiple agents like Codex, Junie, and Claude Agent. This decision was based on a detailed assessment using a benchmark dataset that included real-world software engineering tasks across Java, C#, and Python ecosystems, focusing on three main criteria: solve rate, cost, and latency. Codex, particularly its GPT-5.4-mini model with medium reasoning, outperformed other candidates by achieving the best combination of solve rate and cost-effectiveness, despite having a higher latency in some cases. The evaluation also included an online A/B test with real users to confirm these findings in practical scenarios. The aim is to provide an agent that integrates seamlessly into developers' workflows, although users retain the flexibility to switch to other agents as their needs evolve. As the landscape of AI models and agents continues to develop, JetBrains plans to revisit and potentially update their recommendations based on new data and advancements.