Skip to main content
●Measured savings across 11 LLMs — Claude Opus 4.7 to Gemini Flash.→ See per-model data
or get an API key →
Sign InGet free API key
Connect your client
FeaturedResearchSources: 4

Perfect retrieval isn't enough: the hidden cost of long LLM context

LLMs lose 13.9%–85% accuracy as context grows even with perfect retrieval and full masking. Compression is a lever that improves accuracy and cost at the same time.

2026-06-01James Hollingsworth7 min read read
Read article
Latest posts (3)
2026-05-20Engineering

One Query Parameter Cut 19% of Our Claude Code Context Burn

James Hollingsworth · 7 min read
2026-05-20Research

Why Google Killed CNET's AI Article Strategy (And What We're Doing Instead)

James Hollingsworth · 11 min read
2026-05-20Cost

MCP Token Costs: A Quantitative Breakdown

James Hollingsworth · 9 min read
See all posts in feed

All posts

Engineering, research, cost, and tutorial posts in one feed.

Showing 10 of 44
Date
Title
Category
Author
2026-05-20We Compressed a 3,448-Line MCP Gateway to a Searchable Symbol MapEngineeringJames Hollingsworth→2026-05-19Resend bounce handling, PostHog funnel instrumentation, and an AST fence against billing driftEngineeringJames Hollingsworth→2026-05-09How we measure quality at 5× compression: corpora, judge prompts, and anti-bias controlsResearchJames Hollingsworth→2026-05-09gotcontext + TokenSpeed: stack input compression with TRT-LLM-class inference for self-hostersEngineeringJames Hollingsworth→2026-05-08gotcontext + rtk: stack two compression layers, kill 90%+ of your token billEngineeringJames Hollingsworth→2026-05-08Your LLM Gets Measurably Worse as the Conversation Grows. All of Them Do.ResearchJames Hollingsworth→2026-05-08You Can Cut Chain-of-Thought Token Costs ~66% With One Prompt ChangeEngineeringJames Hollingsworth→2026-05-08Why Long Agent Sessions Fall Apart (And the Paper That Explains It)EngineeringJames Hollingsworth→2026-05-08What Chunking Strategy Actually Matters for RAG QualityResearchJames Hollingsworth→2026-05-08Vision Tokens Are Expensive and Nobody Reads the Pricing PageCostJames Hollingsworth→
Email updates
One engineering post a week,
no marketing in between.

Long-form pieces from the team building gotcontext. Compression internals, model evals, customer post-mortems. RSS feed available — see /blog/rss.xml. Unsubscribe with one click.

Subscribe — one engineering post a weekSubscribing creates a free gotcontext account. You can unsubscribe from emails without canceling the account.