Open Coding Agents, launched by Ai2, aim to democratize access to coding tools for developers by providing efficient, open-source models. The latest release, SERA-14B, adds to a growing family of coding agents designed to adapt to any codebase, making them suitable for tasks like code generation, debugging, and maintenance. The cost of training these models has been reduced significantly, with the ability to replicate high performance at a fraction of the price of previous models—about $400 for basic performance and $12,000 to match top-tier proprietary models. SERA models, particularly SERA-32B, outperform existing models on SWE-Bench tasks while requiring only 40 GPU days for training. They can integrate seamlessly with existing codebases and are optimized for NVIDIA's infrastructure, achieving impressive throughput rates. The soft-verified generation (SVG) technique allows for the creation of synthetic training data that doesn't need to be fully correct, simplifying the training process. This method, combined with a taxonomy of common bug patterns, enables the generation of diverse data sets without the exhaustive testing usually required. The innovations in synthetic data generation allow small businesses and independent developers to fine-tune these coding agents to their specific needs without extensive resources. The approach emphasizes generating data that reflects real developer workflows rather than focusing solely on correctness, making it easier to adapt the models to unique coding environments. Overall, this release represents a significant step towards making powerful coding tools accessible to a broader audience.