LTM Models by Magic

MagicProprietaryCoding

2 models2023–2024Up to 100m ctx

Last refreshed 2026-05-19. Next refresh: weekly.

Details

ResearcherMagic

LicenseProprietary

Commercial useCommercial use: conditional

Models2

Released2023–2024

Max context100m

Links

Website

About

LTM (Long-Term Memory) is Magic's model family focused on extreme-scale context windows for software development. LTM-1 (2023) introduced a 5M-token context; LTM-2-mini (2024) extended this to 100M tokens using a novel sequence-dimension algorithm. Developed by Magic (magic.dev), San Francisco.

Decision facts

Best fit: codingcode
Capability starting point: LTM-2-mini with 100m context
Lowest tracked input: Not tracked
Closest related family: SWE-1

Current Variants

Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.

2 in view

LTM-2-miniCurrent

Use when the workload needs code, 100m context, and coding.

2024-08code100m contextcoding

LTM-1Current

Use when the workload needs code, 5m context, and coding.

2023-06code5m contextcoding

Current LTM variants with use-when guidance and lifecycle status
Model	Use when	Released	Signals	Status
LTM-2-mini	Use when the workload needs code, 100m context, and coding.	2024-08	code100m contextcoding	Current
LTM-1	Use when the workload needs code, 5m context, and coding.	2023-06	code5m contextcoding	Current