Cognition Launches SWE-2 and Raises $2 Billion

Cognition has released its coding model SWE-2, now generally available (GA). The company says it matches recent frontier models on mainstream benchmarks while costing up to 70% less, making it the closest model to frontier-level performance to date. SWE-2 is already available on Devin's desktop app and CLI, and will be free for all Pro, Max, and Teams subscribers over the next month. Alongside the launch, the company closed a $2 billion funding round, and founding team member regulargio officially joined, kicking off a period of密集 releases.

Confirmed

Cognition officially claims SWE-2 matches recent frontier models on mainstream benchmarks at up to 70% lower cost
SWE-2 is available now on Devin desktop and CLI; free for Pro, Max, and Teams users for one month
An official blog post details how SWE-2 was trained
The company raised $2 billion

Technical highlights

As relayed by @andrewncarr, SWE-2 is Cognition's first model with effort tiers; the team made major improvements to their length penalty approach, lifting the Pareto curve within a single RL training run
The team says they scaled reinforcement learning to multi-trillion parameter scale and refined the RL recipe, advancing capability and cost simultaneously
Per @petrusenkomax, SWE-2 claims performance competitive with Fable 5.1 and GPT-Astra — Cognition's direct push into core coding models following its Agent product line

2026-09-10 ~ 2026-09-11 · 5 related posts

Primary sources

Cognition Ships SWE-2 at 70% Lower Cost Alongside $2B Raise — DevinAI · 2026-09-11 00:04 (UTC+08:00)
Cognition ships SWE-2 in Devin, free for Pro, Max & Teams subscribers for a month — Scobleizer · 2026-09-11 01:29 (UTC+08:00)
Cognition Ships SWE-2, Its First Model With Effort Levels, Pushing the Pareto Curve in a Single RL Run — andrew_n_carr · 2026-09-11 01:55 (UTC+08:00)

Cognition launches SWE-2: frontier-level coding performance at up to 70% lower cost — silasalberti · 2026-09-10
[source] Cognition Ships SWE-2 at 70% Lower Cost Alongside $2B Raise — DevinAI · 2026-09-11
Cognition launches SWE-2 coding model, rivaling Fable 5.1 and GPT-Astra — petrusenko_max · 2026-09-11
[source] Cognition ships SWE-2 in Devin, free for Pro, Max & Teams subscribers for a month — Scobleizer · 2026-09-11
[source] Cognition Ships SWE-2, Its First Model With Effort Levels, Pushing the Pareto Curve in a Single RL Run — andrew_n_carr · 2026-09-11