Magic.dev
Frontier coding model with 100M-token context window
Overview
Magic.dev is building frontier coding models specialized for software engineering — reportedly the largest context window in the industry (100M+ tokens) for full-codebase reasoning. Limited public availability; enterprise focus. Backed by Google, Nat Friedman, Daniel Gross.
Use cases
What you can produce with Magic.dev
- Join the waitlist on magic.dev to request early access to Magic's long-context coding models when availability opens up.
- Read the company's published research on 100M-token context windows, including the LTM-2-Mini announcement and its HashHop long-context evaluation method.
- Evaluate the enterprise early-access programme if your organisation wants whole-repository AI reasoning without building retrieval pipelines.
- Follow Magic's blog and X account for model and infrastructure announcements, such as its Google Cloud supercomputer buildout.
- Explore open engineering and research roles at one of the most heavily funded AI coding labs if you work in ML systems or model training.
- Use its public benchmark write-ups to understand why context length matters for codebase-scale tasks before choosing any AI coding tool.
ASEAN Perspective
Magic.dev in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
Magic.dev is an AI research company building frontier code-generation models, best known for pursuing extremely long context windows (its LTM models targeting up to 100M tokens) aimed at reasoning over entire codebases as an autonomous software engineer. The ambition and research are genuinely notable, and the long-context approach addresses a real limitation of mainstream coding assistants.
It suits those tracking the cutting edge of AI software engineering rather than buyers seeking a ready tool today. Caveats: the product has had very limited general availability, so for most users it is more promise than usable product; benchmark-to-real-world performance is unproven at scale; and it competes against well-shipped tools (Cursor, Copilot, Claude Code, Devin). Treat capability scores as provisional given limited public access; no broadly available public API.
What people say
Magic.dev is one of the best-funded and least-shipped companies in AI coding. Founded in 2022 by Eric Steinberger and Sebastian De Ro, it has raised hundreds of millions of dollars — including a $320M round led by Eric Schmidt with participation from CapitalG, Atlassian, Sequoia, Jane Street, Nat Friedman and Daniel Gross — and partnered with Google Cloud to build dedicated NVIDIA-powered supercomputers. Its headline research result, the LTM-2-Mini model announced in 2024, claims a 100-million-token context window: enough to hold roughly ten million lines of code in a single prompt.
That is where the verifiable story largely ends. As of mid-2026 Magic still operates in waitlist and early-access mode with no generally available product. There are no meaningful G2, Capterra or app-store reviews because almost nobody outside the company and select partners has used the models, and independent teardowns have questioned whether revenue matches the valuation. Developer discussion on Hacker News and X is a mix of genuine technical admiration — the LTM architecture's claimed ~1000x efficiency gain over standard attention at extreme context lengths is taken seriously — and growing scepticism about a company approaching its fourth year without a public release.
The pitch, if it lands, is compelling: full-codebase reasoning without retrieval hacks, where the model simply reads your entire repository, documentation and history in context. Competitors like Cursor, Claude Code and Copilot ship incrementally today; Magic is betting everything on a step-change.
Right now Magic.dev fits only two audiences: engineering leaders at large organisations willing to pursue enterprise early access, and people tracking frontier AI research. There is nothing to download, no pricing page, and no user community. Treat it as a research lab to watch rather than a tool to adopt, and re-evaluate if a public product actually ships.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Magic.dev's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Magic.dev unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Magic.dev directly →
Spotted something out of date? Suggest an update →
More in Code & Dev Tools
