Cognition Raises $2 Billion at a $48 Billion Valuation, Closing In on Cursor
Set Trending Topics as a preferred source on Google.
Cognition, the U.S. startup behind the AI coding agent Devin, has raised more than $2 billion in a Series E round. The deal values the company at $48 billion. For comparison, SpaceX paid $60 billion in stock for Cursor parent Anysphere. The two most important players in AI-assisted software development are now only about a quarter apart on price.
Who Invested
The round was led by Andreessen Horowitz and Accel, both new to the cap table. Existing backers Founders Fund, General Catalyst and Avenir also joined. The list of further participants is unusually long and includes Benchmark, Bessemer, Kleiner Perkins, Greylock, Lightspeed, Altimeter, Bond Capital, Meritech, T. Rowe Price, Lux Capital, 8VC, DST, Bain Capital Ventures, Ribbit, Battery, Stripes, Alpha Wave and Alkeon. In total, Cognition names more than 30 funds in this round.
The repricing has been steep. In its previous financing round in the spring, Cognition was valued at $26 billion. The company, founded in 2024, has therefore nearly doubled in value within a few months.
Revenue Approaches the $1 Billion Mark
Cognition puts its annualized run rate at close to $900 million. At the time of the spring round, that figure stood at $492 million, which amounts to almost a doubling in roughly four months. The metric projects current revenue over a full year, so it works as an industry benchmark rather than as audited annual revenue.
Among its reference customers, the company names NVIDIA in chip design, GE Aerospace in aviation, Citi in financial services, Mercedes-Benz in the automotive sector and the AI infrastructure provider Modal.
What Sets Devin Apart From Code Assistants
Devin is positioned as an autonomous software engineer, a different design goal from autocomplete tools inside the editor. Cognition’s core idea: engineers should work more like architects and delegate execution to swarms of agents. The company has recently added several features that point in that direction:
- Auto-Triage: Devin takes a first pass at investigating incidents.
- Security Swarm: Multiple agents find and triage vulnerabilities in the codebase.
- Automations: Tasks start automatically from events in Slack, GitHub, Linear or other systems, so teams can skip opening a chat for every task.
Cognition describes its long-term goal as “self-driving software”: agents that act proactively, software that improves itself, and compute budgets that shift on their own toward the highest impact use cases.
Strategically, Cognition positions itself as an independent agent lab. The company wants to combine models depending on the job, including its own, keeping customers free of any single provider. In practice, Devin currently has both OpenAI’s GPT-6 Astra and Anthropic’s Claude Fable 5.1 integrated. Cognition itself points out that on its in-house FrontierCode 1.1 benchmark, Astra lands within 0.4 points of Fable 5 at 64 percent lower cost. Trade-offs like that, between quality and cost per task, are the argument for a multi-model setup.
There are in-house models as well. With the SWE family, Cognition is building a line of LLMs tailored specifically to software engineering. The current SWE-1.7 was introduced in the summer and is also used inside Devin. One caveat applies: the base is Kimi K2.7, which Cognition trained further with its own reinforcement learning for complex programming tasks. So “in-house LLM” holds in the sense of a model the company developed further itself, while the pretraining came from elsewhere.
In line with its aim of working closely with customer systems, Cognition has expanded its office network considerably. Over the past year, the hubs in San Francisco, New York and Austin were joined by offices in Washington, D.C., Tokyo, Singapore, London, São Paulo and Madrid.
