SWE-2
A next-generation code intelligence model released by Cognition, optimized specifically for software engineering tasks. It is designed to provide end-to-end software engineering capabilities, including understanding requirements, locating bugs, modifying codebases, and passing test suites.
Timeline (last 90 days)
Cognition, the company behind Devon, is valued at $48 billion, surpassed $1 billion in annualized revenue in September, and runs its own model SWE-2.
SWE-2 is a coding model built by Cognition on top of Moonshot AI's Kimi K2 model, released on September 10, with a primary focus on cost efficiency
Devin's pricing page states that SWE-2 is free to use in its desktop application and command line until October 16
Cognition states that SWE-2 scores only one point below Anthropic's Claude model on its own coding benchmarks, yet costs 64% less
Cognition has launched a new-generation programming model, SWE-2.
SWE-2 is post-trained on Kimi K3 and uses reinforcement learning (RL), optimizing along both the cost and capability dimensions simultaneously.
SWE-2 has been integrated into Devin Desktop and CLI, and users can use it directly.
SWE-2 achieved a score of 50.0% on the FrontierCode 1.1 Main benchmark.
SWE-2's benchmark score differs from that of competitor Fable 5.1 by less than one percentage point, but its cost is 64% lower.
Compared to GPT-6 Astra, SWE-2's score trails by a few percentage points, but its cost is only one-quarter of the latter's.
15 more timeline events
All Facts (20)
Cognition, the company behind Devon, is valued at $48 billion, surpassed $1 billion in annualized revenue in September, and runs its own model SWE-2.
50%UnverifiedCognition states that SWE-2 scores only one point below Anthropic's Claude model on its own coding benchmarks, yet costs 64% less
50%UnverifiedSWE-2 is post-trained on Kimi K3 and uses reinforcement learning (RL), optimizing along both the cost and capability dimensions simultaneously.
50%UnverifiedSWE-2 has been integrated into Devin Desktop and CLI, and users can use it directly.
50%UnverifiedSWE-2 achieved a score of 50.0% on the FrontierCode 1.1 Main benchmark.
50%UnverifiedSWE-2's benchmark score differs from that of competitor Fable 5.1 by less than one percentage point, but its cost is 64% lower.
50%UnverifiedCompared to GPT-6 Astra, SWE-2's score trails by a few percentage points, but its cost is only one-quarter of the latter's.
50%UnverifiedWhen completing the same tasks, SWE-2 requires 58% fewer interaction turns than its predecessor SWE-1.7, costs 81% less, and achieves a higher score.
50%UnverifiedSWE-2 takes the pragmatic route of maximizing cost advantage within a sufficiently good performance range, rather than pursuing absolute top performance.
50%UnverifiedThe naming of SWE-2 points to a capability upgrade in the direction of Software Engineering.
50%UnverifiedSWE-2 is a newly released coding model from Cognition, an iterative version of the Devin engine, specifically trained and fine-tuned for software engineering scenarios.
50%UnverifiedDevin Voice's capabilities are supported by two core technologies, GPT-Live and SWE-2, responsible for voice understanding and code generation respectively.
50%UnverifiedSWE-2 represents the autonomous agent approach, in which the system independently plans, executes, and verifies after receiving a task objective, with humans primarily responsible for issuing tasks and reviewing final results.
50%UnverifiedSWE-2 is very likely a model specifically optimized for software engineering tasks, rather than a simple variant of a general-purpose large language model.
50%UnverifiedCognition officially claims that SWE-2 is capable of directly competing with top-tier models such as Fable 5.1 and GPT-Astra.
50%UnverifiedSWE-2 is a coding model built by Cognition on top of Moonshot AI's Kimi K2 model, released on September 10, with a primary focus on cost efficiency
50%UnverifiedCognition has launched a new-generation programming model, SWE-2.
50%UnverifiedThe Cognition team released SWE-2.
50%UnverifiedCognition has released SWE-2, a new generation software engineering agent.
50%UnverifiedCognition has launched a new generation code intelligence model, SWE-2.
50%