NSA News

YES

Some variation of NSA (Native Sparse Attention)

>=25T pretraining tokens

>= 2 shared experts

>=52B active parameters

>=16 active experts

uses some non-AdamW optimizer

>=512 experts

1M+ Context

intra-expert communication

Some image input (multimodality)

>=1.5T parameters

DS-MoE with adaptative expert count

>= 50% on TerminalBench (https://www.tbench.ai/leaderboard)

>= 73% on SWE-Bench Verified (according to epoch.ai)

Gemini 2.5 Pro tier or higher on FictionBench (90.6%+ at 192k)

>= 60% on BrowseComp (https://www.kaggle.com/benchmarks/openai/browsecomp)

DeepSeek reports some results with a full-blown deep research agent, and emphasizes that this is the intended use-mode

>= 44% on Humanity's Last Exam (text only) at scale.com leaderboard

YES

Chief of Staff to the President

NASA administrator

Senate Majority Leader

Senate Minority Leader

Director General of the BBC (british broadcasting corporation)

USA President

Secretary General of the UN

NSA director

Secretary of the Treasury

CEO of Google

FBI director

Chief Justice of Supreme Court

Chairman/Chairperson of the Joint Chiefs of Staff

US Secretary of Defense

Chief of Staff of the Army

Head Coach of a Major League Baseball team

Cardinal in the Catholic Church

no

DOC

DHS

Other

FDA

CDC

DOE

NSA

ICE

DEA

SEC

CIA

FBI

DOD

FCC

DOJ

Department of Commerce

Federal Emergency Management Agency (FEMA)

Central Intelligence Agency (CIA)

National Security Agency (NSA)

Federal Bureau of Investigation (FBI)

Department of Homeland Security

Internal Revenue Service (IRS)

Securities and Exchange Commission (SEC)

Department of Interior

Department of Defense

Department of State

Department of Energy

NO

YES

YES

NO