The context window just jumped to 50 million. a memory layer called galahad-kv, out oct 7, runs…
the context window just jumped to 50 million. a memory layer called galahad-kv, out oct 7, runs that many tokens on a single h100 by caching kv blocks to local nvme instead of recomputing.
the long-read now costs disk space, not gpus.
Context
Verified dates: the paper 'Real Long-Term Memory for AI: A 50-Million-Token Window That Is Faster and Cheaper Than Recompute' is arXiv 2610.10845, dated Oct 7, 2026 on Papers with Code, and was picked up as a Hugging Face daily paper on Oct 8. It describes a memory layer released as the public package galahad-kv and reports measurements on a corpus of 50,000,000 tokens.
The abstract says the layer saves the KV state of each block of about 16,000 tokens to encrypted local NVMe disk and loads it back later, byte-exact, without recomputing it. The PyPI page lists galahad-kv 1.31.0 uploaded Oct 1, 2026, a KV-cache layer for vLLM and SGLang.
The Oct 7 date, the 50 million token corpus and the NVMe block caching match the paper. The single H100 claim and the line that the long read costs disk space, not GPUs, were not seen in the sources read, so unsupported here, not refuted. The 50 million figure is the size of the corpus measured, not a model context window that attention covers at once. A separate Corbenic AI paper (arXiv 2609.39358) by Sietse Schelpe describes the same Galahad memory as making reading a one-time cost. 'The long-read now costs disk space, not GPUs' is the author's opinion.
Related work
- arXiv: Working around the compute ceiling, Galahad's byte-exact memory ↗Corbenic AI, October 2026.
- cere-bro: galahad-kv, a 50M-token KV memory on NVMe ↗Oct 9, 2026; written from the abstract.
- USENIX OSDI 2026: no buffer, no bottleneck, zero-copy KV cache offloading ↗Related work on KV offloading.
Watch next
- Find the GPU used in the paper's measurements.
Sources
- arXiv: Real Long-Term Memory for AI, a 50-million-token windowarxiv.org
- Papers with Code: Real Long-Term Memory for AIpaperswithcode.co
- PyPI: galahad-kv 1.31.0pypi.org
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 11 October 2026 at 17:01 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →