Research code models intended to reason over large repositories, documentation and libraries.
Magic Funding History
Magic develops models for software engineering and AI research. Its work combines long context, model pretraining and reinforcement learning, with the aim of moving from short code suggestions to larger engineering tasks.
Company Overview
Magic
Founded 2022 · San Francisco, California, US
Official WebsiteMagic Funding Timeline
Financing
$320M
Magic announced US$320 million in new investment without specifying a lead investor or a Series label.
Participation included Eric Schmidt, Jane Street, Sequoia Capital, Atlassian, Nat Friedman, Daniel Gross, Elad Gil, CapitalG.
Source · magic.dev ↗Financing
$117M
Magic identified Nat Friedman and Daniel Gross as leading this US$117 million financing. No valuation was disclosed in the announcement.
Led by Nat Friedman, Daniel Gross. Participation included CapitalG, Elad Gil.
Source · linkedin.com ↗Series A
$23M
Magic disclosed this Series A in 2023-02-06. The amount records this financing, rather than the company valuation.
Led by CapitalG. Participation included Amplify Partners, Nat Friedman, Elad Gil.
Source · prnewswire.com ↗Seed
$5M
Disclosed retrospectively as occurring the previous summer; no exact date or lead is assigned.
Source · magic.dev ↗
Magic Founding Team
Magic’s founders combine model research with a systems-heavy engineering program. The company recruits across pretraining, reinforcement learning, kernels and supercomputing.
Magic Products
Magic Company Development
Financing a Model Lab, Not Just an Editor
Eric Steinberger and Sebastian De Ro founded Magic around an AI software-engineer ambition. The February 2023 announcement disclosed a $23 million Series A following a $5 million seed round. CapitalG led the Series A. These announcements positioned Magic as a company developing its own models rather than simply adding a chat interface to another provider’s API.
Long Context Becomes the Technical Direction
In August 2024, Magic described LTM-2-mini research on a 100-million-token context window and its HashHop evaluation. It also announced a $320 million investment and plans for Google Cloud supercomputers using NVIDIA hardware. The strategic connection was explicit: reasoning over large codebases and libraries requires both model changes and substantial training and inference infrastructure.
Efficiency Joins the Scaling Program
A September 2026 research update shifted attention to compute-efficient pretraining and larger models. Magic describes its results as internal research claims rather than a generally available coding assistant release. Its current website reports US$768 million raised in total without a dated breakdown of that cumulative figure.
FAQ
What does Magic build?
Magic researches code models for software engineering and AI research, including models designed to use very large amounts of project context.
Is Magic a code editor?
Its public materials emphasize model and infrastructure research. They do not establish a generally available editor product.
What is LTM at Magic?
LTM is Magic’s Long-Term Memory model line. Its 2024 research update discussed an experimental 100-million-token context window.
How much funding has Magic announced?
The current company website reports $768 million raised in total. Separately disclosed financings include the 2023 Series A, a $117 million 2024 financing, and the August 2024 $320 million investment.
Who founded Magic?
Eric Steinberger and Sebastian De Ro are identified in the company’s Series A announcement.