Cognition Releases SWE-2 Coding Agent with Near-Flagship Performance at 64 Percent Lower Cost
Cognition has launched SWE-2, its latest software engineering-focused model developed through post-training on the 2.8-trillion-parameter Kimi K3 base. By applying a proprietary reinforcement learning pipeline scaled to trillion-parameter models, Cognition achieved a 5 to 6 percentage point increase in benchmark accuracy. This approach allows SWE-2 to record a 50.0% score on the FrontierCode 1.1 Main benchmark, outperforming older models like SWE-1.7 and Grok 4.6 in both quality and processing cost.
Related tools
Recommended tools for this topic
These picks prioritize high-intent tools relevant to this topic. Some links may include partner or affiliate tracking.
Strong fit for AI, backend, and frontend readers looking for an AI-first coding workflow.
View CursorNatural next step for readers evaluating LLM adoption, APIs, and production inference.
Explore APIA strong fit for readers comparing Claude-class models, safety, and long-context workflows.
View AnthropicComparison
| Aspect | Before / Alternative | After / This |
|---|---|---|
| Base Model and Scale | SWE-1.7 infrastructure / older base models | Kimi K3 (2.8T parameters) with custom RL pipeline |
| FrontierCode 1.1 Score | Lower efficiency, outpaced by Fable 5.1 and GPT-5.6 Sol | 50.0% score, nearly matching Fable 5.1 |
| Inference Cost | High premium rates for flagship model performance | 64% cost reduction compared to Fable 5.1 |
| Cost vs GPT-6 Astra | High cost ceiling for maximum performance | Near-equivalent capabilities at 1/4 of the cost |
Source: Hacker News
This page summarizes the original source. Check the source for full details.
