Purpose: According to the official text, Magic positions itself as an AI company committed to building safe AGI to accelerate humanity's progress on the world's most important problems. It is building frontier code models to automate software engineering and research. Magic believes that the most promising path to safe AGI is to automate AI research and code generation, thereby improving models and solving alignment problems more reliably than humans can do independently.
Main capabilities: The official text states that its approach combines frontier-scale pretraining, domain-specific reinforcement learning, ultra-long context, and inference-time compute. In addition, the official text mentions a research update: >10x More Efficient Pretraining, that is, research on compute-efficient pretraining and scaling to trillion-parameter models. The official text also states that the company has thousands of GB200 compute units.
How to use: The official text does not provide user-facing product usage instructions, installation commands, code examples, or specific operating steps. Therefore, the specific way to use Magic is not yet provided.
Applicable scenarios: The official text is mainly oriented toward directions such as automated software engineering, AI research, model improvement, and alignment research. Its job openings also center on research engineering, inference and RL systems, kernels, pretraining systems, RL research and environments, safety engineering, and so on. Regarding the product forms that ordinary users or developers can directly use, the official text does not explain.
Limitations and information not provided: The official text does not provide pricing model, price, platform support, product feature details, performance data, or detailed usage steps. This information is not yet provided.
Funding and team: Magic has raised $768 million from Nat Friedman, Daniel Gross, CapitalG, Elad Gil, Sequoia, Jane Street, Eric Schmidt, and others. The company describes itself as a small group of engineers and researchers working on the few foundational research problems that lead directly to AGI, and welcomes relevant talent to apply.
