Article
CommunityJev Guides & Articles
The Year in AI Papers
A browsable atlas of 1,018 AI papers from August 2025 to August 2026, sorted into 24 topic collections with labs, citations, and code links, plus a reproducible cost benchmark for summarizing 1,000 papers.
- Format
- Article
- Reading time
- 3 min
- Published by
- Community
- Last verified
- Sep 2026
Highlights
- The corpus is 1,000 frozen papers from August 4, 2025 to August 4, 2026 plus 18 verified Together AI papers, for 1,018 total.
- Topic areas range from Reasoning (177 papers) and Agents (275) to Video and spatial AI (164), Robotics (59), and Safety (35).
- The 1,000 papers span 30,681 pages and 102.7 million extracted characters; oversized ones use a 50,000-character map-reduce pass.
- Cost benchmark: DeepSeek V4 Flash summarized all 1,000 papers for $3.99 total ($0.004 each), versus $6.00 for GPT-5.6 Luna and $35.76 for Claude Haiku 4.5.
- The authors call it a popularity-weighted field guide and cost benchmark, not an exhaustive history or a factual-quality ranking.
Watch out
Neither the site nor its Nutlope/1kpapers repository documents Jev - the published classify-topics script calls Together AI models - and the benchmark measures inference cost only, not factual quality.
Questions about this entry
Short answers drawn from the linked project or page.
Nothing in the fetched page or its repository mentions Jev, and the published classify-topics script calls Together AI models. So there is no integration to inspect here: this is a research atlas that happens to be listed. Its one measurable claim, the cost of summarizing 1,000 papers, is about DeepSeek V4 Flash, GPT-5.6 Luna and Claude Haiku 4.5, not about Jev.
The page makes a real effort to hold it constant: same extracted text, same chunk boundaries, same prompt, same final-summary contract, reasoning disabled, and provider-reported token counts with prices frozen on a stated date. DeepSeek V4 Flash came to $3.99 for 1,000 papers at $0.004 each, GPT-5.6 Luna $6.00, Claude Haiku 4.5 $35.76. What is excluded is everything outside inference, including PDF extraction, storage, networking and dev time.
No, and the page says so plainly: it is a cost comparison, not a factual-quality ranking. It also disclaims being an exhaustive history, describing itself as a reproducible, popularity-weighted field guide selected for citation impact, official-code adoption, recency and field-wide significance. Paper selection is therefore a popularity proxy, so a missing or obscure paper is a property of the method, not an oversight.
From 8,262 discovery candidates sourced from Hugging Face Daily Papers plus verified official research from five frontier labs, deduplicated by canonical arXiv ID, with every version-pinned PDF checked and exactly 1,000 papers frozen from August 4, 2025 to August 4, 2026. That is 30,681 pages and 102.7 million extracted characters, and oversized papers go through a deterministic 50,000-character map-reduce rather than silent truncation. 18 verified Together AI papers bring the atlas to 1,018.
By month, by research lab, or by topic across 24 collections. The labs listed are OpenAI, Anthropic, Moonshot AI, DeepSeek, MiniMax and Z.ai / GLM. The topics are dominated by Agents at 275 papers, Reasoning at 177, Video and spatial AI at 164, Multimodal at 144, Systems at 143 and Robotics at 59, down to Safety at 35. There is also a most-cited view and a most-starred view.
Just this set, and worth reading that way. The month strip we captured runs from 63 papers in August 2025 up to 115 in May 2026, with 100 in March 2026, 98 in October 2025, 84 in July 2026 and 3 in August 2026 for the partial current month. Those are counts of the frozen 1,018, not of everything published in each month.
The captured front page states none for either, so do not assume any. It describes 1,000+ curated and indexed papers and carries the cost table, but the license terms live in the Nutlope/1kpapers repository rather than on the page. Check the repository directly before reusing the selection or the extracted text.
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.