GoKawiilCognition releases SWE-2 coding model built on Kimi K3, undercutting rivals on price
Cognition unveiled SWE-2, a coding-focused AI model post-trained from the 2.8-trillion-parameter Kimi K3 base. The company says it scores 50.0% on FrontierCode 1.1 Main, nearly matching Fable 5.1 while costing 64% less, and comes close to GPT-6 Astra at roughly a quarter of that model's price. SWE-2 also outperforms Cognition's earlier SWE-1.7 model and Grok 4.6 across several coding benchmarks.
GoKawiil's interpretation of the reporting above, not reported fact.
The release shows AI labs increasingly competing on cost-efficiency rather than just raw benchmark scores, since coding models are widely used in production developer tools where inference costs add up quickly. By training multiple reasoning-effort levels in a single reinforcement learning run, Cognition claims a more efficient path to improving models built on top of existing large base models like Kimi K3, rather than training from scratch.
- SWE-2 scores 50.0% on FrontierCode 1.1 Main, near Fable 5.1's 50.9% but at 64% lower cost.
- The model is post-trained from Kimi K3, a 2.8-trillion-parameter base model, using a new RL technique across reasoning-effort levels.
- SWE-2 outperforms Cognition's prior SWE-1.7 model and Grok 4.6 on multiple coding benchmarks while trailing GPT-6 Astra by only a few points.
Keychron K2 Wireless Mechanical Keyboard — If you're going to spend your day pair-programming with an agentic coding model like SWE-2, the keyboard under your hands matters. The Keychron K2 gives you hot-swappable mechanical switches in a compact 75% layout with Mac and Windows support, plus Bluetooth for hopping between machines. It's a favorite among developers who live in the terminal and editor.
See Keychron K2 Wireless Mechanical Keyboard on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: cognition.com, 2026-09-10
Published there as: “Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.