Overview
1kpapers (1kpapers.com) is Hassan El Mghari's (@nutlope) live atlas of AI research papers. A September 2026 clip describes classifying one thousand eighteen papers for about eight cents total with roughly two hundred fifty six millisecond median end-to-end latency per paper. DeepSeek V4 Flash summarizes each PDF; Jev Choice picks one of twenty four topics from title plus summary. The site may still show editorial topics while experimental Jev labels are evaluated, per the builder thread.
Problem: Research atlases need stable topic labels across hundreds of papers without paying frontier-model prices per row or waiting on slow batch jobs.
Built for: ML researchers, technical writers, and builders curating large paper sets who want cheap, fast topic routing with typed Jev labels.
First indexed on Jev Directory: 2026-09-21
Creator and team
- Name
- Hassan El Mghari
- Organization
- Together
- Handle
- @nutlope
“I used Jev to classify 1,018 AI research papers. The result: $0.08 total cost and 256ms median end-to-end latency per paper.”
How Jev is used
- Role in the product flow
- Topic Choice after cheap summarization on each paper row
- Primitives
- Choice
- State in
- Paper title plus DeepSeek V4 Flash summary text and a fixed menu of twenty four topic labels (per launch clip).
- Decision out
- Single topic Choice per paper for atlas filtering and map placement.
- Summarize each paper with DeepSeek V4 Flash
- Send title, summary, and twenty four topic options to Jev
- Jev Choice selects the best-matching topic
- Publish labels to the 1kpapers atlas UI
1kpapers separates cheap text generation from typed routing. Summaries can drift; Jev Choice forces each paper into one of twenty four explicit topics so the atlas stays sortable and filterable without embedding indexes. Parallel batches keep median latency near a quarter second per row at micro-dollar prices in the attributed clip. Hassan noted the Jev labels are still experimental: readers may see editorial topics on the live site while evals run. Treat the X thread and 1kpapers.com as canonical for current behavior, not a frozen schema export.
Sourced performance claims
- One thousand eighteen papers classified for about eight cents total; about two hundred fifty six millisecond median latency per paper.Source: nutlope X post 2100426999546184123
- Twenty four candidate topics per classification call.Source: nutlope X post pipeline description
Features and stack
Features
- Live paper atlas at 1kpapers.com
- DeepSeek V4 Flash summaries plus Jev topic Choice
- Sub-dollar batch cost on one thousand plus papers in the clip
- Featured showcase video from @nutlope
Stack
- 1kpapers.com
- DeepSeek V4 Flash
- TypeSafe System One
Pricing: Public atlas browsing is free; rebuilding the pipeline spends summarization and System One tokens per paper.
Demo
Showcase clip with the same lightbox player as the homepage. Click to play.
1kpapers atlas with Jev topic Choice
One thousand eighteen papers, eight cents, two hundred fifty six milliseconds median. DeepSeek summarizes; Jev picks the topic so your research map is not hashtag soup.
Links
FAQ
- Are Jev topics live on every paper card?
- Hassan's thread says Jev labels are experimental. The site may still display editorial topics while evaluations finish. Refresh 1kpapers.com for the latest UI.
- Why Choice instead of embeddings?
- The launch pipeline sends explicit topic strings to Jev Choice after summarization, avoiding vector indexes for the first pass at atlas scale.
- Can I reproduce the eight cent figure?
- It comes from the September 2026 clip on one thousand eighteen papers. Re-run on your corpus with current Together and TypeSafe pricing for production budgeting.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- Jev use cases
The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.
- Jev vs LLM classification
When to gate with System One probabilities instead of asking a chat model to label things.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.