Magic.dev
Coding-model research lab; nothing released yet
Overview
Magic (Magic AI, Inc.) is a San Francisco lab training code models for autonomous software engineering. It announced a 100M-token-context model, LTM-2-mini, in 2024 and a pretraining update in September 2026, but has released no product, API or pricing. It says it has raised $768 million.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 12 Oct 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- No waitlist, API or sales contact on magic.dev
- Research posts and the HashHop eval are public
Use cases
What you can produce with Magic.dev
- The 100M-token context post, covering LTM-2-mini's design and its cost comparison with Llama 3.1 405B attention
- The open HashHop eval on Magic's GitHub, for testing long-context retrieval without semantic hints
- The September 2026 pretraining post, with its scaling-law method and held-out evaluation design
- Magic's safety page on alignment priorities and AI-assisted alignment research
- A vulnerability-disclosure programme for reporting security issues
Magic pursues ultra-long context, with LTM-2-mini trained on up to 100 million tokens, and by its own account compute-efficient pretraining, aimed at automating software engineering. The research write-ups are substantive and worth reading.
There is nothing to use. Its September 2026 post still says "we look forward to releasing the thing!", and the site has no waitlist, API or price. Its efficiency claims are self-reported and unreplicated. Follow it as a lab; choose a shipping tool for actual work.
What we found
Magic is a San Francisco research lab, operated by Magic AI, Inc., that has raised a great deal of money and still sells nothing. Its homepage says it has "thousands of GB200s" and has raised $768 million from investors including Nat Friedman, Daniel Gross, CapitalG, Elad Gil, Sequoia, Jane Street and Eric Schmidt. There is no product page, no pricing, no waitlist and no public API; the site carries a blog, a safety statement and job listings. Earlier versions of this page described a waitlist and an enterprise early-access programme, neither of which exists.
Eric Steinberger and Sebastian De Ro co-founded it in 2022, and a Series A followed in February 2023. In June 2023 it described LTM-1, a coding model with a 5-million-token context window. In August 2024 it announced LTM-2-mini, trained on up to 100 million tokens of context, which it equates to about ten million lines of code. Magic says its sequence mechanism is roughly 1,000 times cheaper per decoded token than the attention in Llama 3.1 405B at that length. It also introduced HashHop, a long-context test built on random hashes, and a Google Cloud partnership for GB200 supercomputers. That post put total funding at $515 million, including $320 million from new and existing investors. TechCrunch reported at the time that the company had about two dozen staff and no revenue to speak of.
The latest update, dated 8 September 2026, is about pretraining rather than a launch. Magic claims its recipe matches DeepSeek V4 Pro Base on roughly 50 times fewer FLOPs, measured on its own held-out code, research-paper and maths sets. These are self-reported results with no outside replication. The post ends: "we look forward to releasing the thing!"
So there is nothing to evaluate in use, and the efficiency figures are Magic's alone. Its blog is worth reading if you follow long-context research, but anyone choosing a coding assistant this year will need one that already ships.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
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 →
Alternatives to Magic.dev
More in Code & Dev Tools
