Back to News
Cognition•September 10, 2026
SWE-2
Featured on Blog
Paid
AI Coding Agent
Reinforcement Learning
Software Engineering
Autonomous Agents
Featured
🔥 This release made it to our blog
Cognition Releases SWE-2 Coding Model Based on Kimi K3
Cognition has released SWE-2, post-training Kimi K3 to match leading coding benchmarks while lowering inference costs across multiple effort levels.
Read the Full Story 4 min read
### TL;DR
SWE-2 is an advanced agentic coding model designed to push the Pareto frontier of capability and cost-efficiency. It utilizes a novel reinforcement learning (RL) algorithm that trains across multiple reasoning-effort levels in a single run, enabling high-performance software engineering tasks at a fraction of the cost of previous frontier models.
Key Insights & Metrics
Pricing
Available via Devin platform subscription
Cost structure
Version
2.0
Current release version
Hardware
Cloud-based; requires access to Cognition's infrastructure via Devin Desktop, CLI, or Web
Compute requirements
Category
Paid
Licensing model
Region
United States
Primary region
Key Features
- Multi-trillion-parameter RL training
- Pareto-informed cost penalties for optimized effort-level scaling
- Length-weighted reward baseline for training stability
- Integrated online draft-model training for high-throughput inference
- Focused exploration capabilities reducing redundant steps and costs
Ad
Master AI Marketing: Work Smarter, Not Harder
Unlock the power of AI to automate workflows and scale your results. Join 1.5M+ professionals and stay ahead of the curve with the latest AI insights.
Sponsored
Discussion
0
Upvotes
0
Downvotes
0 reviews
Sign in to leave a review
Reviews
No reviews yet. Be the first to review!