The cost war moved in 12 months from which model to how do we fit 4m tokens into a 200k window and…
the cost war moved in 12 months from which model to how do we fit 4m tokens into a 200k window and shopify gisting is the first one to admit it out loud.
learned tokens for the system prompt beat any 75% cache cut, every single time.
Context
Shopify Engineering's post of 19 August 2026 describes gisting, which compresses context into a set of learned tokens. For its Sidekick GraphQL agent, the system prompt went from about 6,000 tokens to about 1,500, a 4 to 1 ratio, without losing prediction quality. At 350 requests per minute, time to first token went from 438 ms to 354 ms, end to end from 6.8 s to 4.2 s, and throughput from 20.2 to 23.4 queries per second.
The post says gisting and prefix caching are not mutually exclusive.
The 4M into 200k tokens figure, the 75 percent cache cut and beating caching every single time are not in the post. Because the post treats the two as complementary, a cost comparison between them is untested here, not refuted. Being the first to admit it, and the 12 month shift, are the author's opinion. Shopify's example is one agent's system prompt, not 4M token compression.
Related work
- Gisting for LLM performance (InfoQ) ↗Secondary coverage; the URL names Spotify but the text says Shopify.
Watch next
- Follow-up posts from Shopify and other labs on prompt compression.
Sources
- Gisting (Shopify Engineering)shopify.engineering
- Gisting: compressing LLM agent context (ZenML LLMOps database)zenml.io
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 4 September 2026 at 21:04 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 →