Agents' Last Exam research paper by UC Berkeley, 2026

UC Berkeley · Jun 3, 2026 · Agents and evaluation · 15 citations · 393 upvotes

Over 1,000 real, checkable professional tasks written with 250+ industry experts; the hardest tier is far from solved.

Summarised by hand from the abstract.

More from UC Berkeley

Date	Paper	Topic	Citations
Sep 12, 2023	Efficient Memory Management for Large Language Model Serving with PagedAttentionvLLM manages the KV cache like pages of virtual memory, serving models with 2 to 4 times the throughput.Inference and efficiency · Sep 2023	Inference and efficiency	8,469

Topic

Date	Paper	Lab	Citations
Mar 12, 2026	Strategic Navigation or Stochastic Search? How Agents and Humans Reason Over Document CollectionsOn 2,250 questions over 800 PDFs, the best agents match human accuracy but lean on brute-force search rather than planning.Snowflake · Mar 2026Snowflake	Snowflake	1
Sep 21, 2026	RRSI: Regularized Recursive Self-Improvement of Agent HarnessesRRSI keeps self-improving agent harnesses from memorising their training tasks, so gains carry over to new benchmarks.Google · Sep 2026Google	Google	0

About this paper

Authors: Yiyou Sun, Xinyang Han, Weichen Zhang and 305 more
arXiv: 2606.05405 · PDF
Venue: arXiv.org
Citations: 15, 1 influential · Semantic Scholar
Upvotes: 393 · Hugging Face
Code: github.com/rdi-berkeley/agents-last-exam
Lab: UC Berkeley

Changes

Date	What changed
Sep 24, 2026	New paperAdded to the listSep 24, 2026

Sources: each lab's own papers and arXiv, with citation and upvote counts from Semantic Scholar and Hugging Face. One-line summaries are for orientation, not a substitute for the paper. Logos via logo.dev; trademarks belong to their owners.

What it shows

More from UC Berkeley

Other agents and evaluation papers

Changes

New papers by email