Cognition

9 stories

Cognition is an AI coding company founded in 2024 by Scott Wu. Recent coverage has focused on its funding and products: it raised $2 billion at a $48 billion valuation, with annualized revenue rising from $492 million to $900 million since May. It also released the SWE-2 coding model, became the first production customer on NVIDIA Vera Rubin systems at CoreWeave, and was involved in a dispute with rival Factory.

Related topics

38 Companies Join Crunchbase Unicorn Board In September, Led By Semiconductor, Fintech And Space Tech

A total of 38 companies joined the Crunchbase Unicorn Board in September, adding $71B in value. Semiconductors, financial services, and space technology led, with 21 US-headquartered companies and 6 from China. Through September, 311 companies have joined this year, on track for the second-highest year of unicorn creation after 2021. Cognition's valuation rose to $48B, Crusoe to $31B, and Mistral to $24.4B.

Crunchbase News ·

NVIDIA and CoreWeave Partner for Agentic AI Production

At CoreWeave Fully Connected, CoreWeave announced production availability of NVIDIA Vera Rubin NVL72 systems with Spectrum-X Ethernet networking, with Cognition as the first customer running production workloads. Early tests showed Vera Rubin achieving 4.8x higher token throughput versus GB200 NVL72. CoreWeave also launched CoreWeave Forge and plans to offer NVIDIA Vera, a CPU built for AI agents.

NVIDIA Blog ·

Meta's Muse launches on Mac with agentic actions

Meta's AI assistant Muse is now available on Mac, where it can work with files, messages, calendar, notes and mail inside native apps, with opt-in access and approval required for sensitive actions. After launching on mobile and web earlier this month and topping the U.S. App Store, the Mac app deepens competition among consumer AI agents.

TechCrunch ·

Cognition Unveils SWE-2 Coding Model

Cognition launched SWE-2, a coding model scoring 50.0 on FrontierCode 1.1 Main—within one point of Fable 5.1 while 64% cheaper; it was post-trained from the 2.8T-parameter Kimi K3 and uses single-run RL across reasoning levels to improve cost-performance.

Hacker News ·
That is everything