< BACK TO NEWS
importantSYS.SOURCE: Cognition• 2026-09-10T15:29:47Z

Cognition Introduces SWE-2 Coding Model, Competing with Fable 5.1 and GPT-Astra

#ai#coding#cost efficiency#reinforcement learning

Cognition introduces SWE-2, a coding model achieving 50.0% on FrontierCode 1.1 Main while being 64% cheaper than Fable 5.1, leveraging reinforcement learning to optimize cost-performance tradeoffs. The model improves efficiency through focused exploration, reduced steps, and enhanced verification discipline compared to previous versions.

Comments

Read original article

*** END OF TRANSMISSION ***

> META'S AI AGENT MUSE RISES TO NO. 2 ON U.S. APP STORE CHARTS> AI-GENERATED WEB DESIGN AND THE EROSION OF CREATIVE INNOVATION> PROXIMA FUSION INVESTS €140M IN HIGH-TEMPERATURE SUPERCONDUCTING TAPE FOR FUSION REACTORS> AMAZON ENHANCES PRIME VIDEO INTEGRATION WITH SEAMLESS SHOPPING FEATURES> ANTHROPIC'S RESEARCH SHOWS AI AGENTS STRUGGLE WITH CAPTCHA CHALLENGES DURING UNAUTHORIZED ACCESS ATTEMPTS> COMPREHENSIVE ANALYSIS OF 200 ANDROID VULNERABILITIES, BROWSER-PHISHING CAMPAIGNS, AND 119K FAKE E-COMMERCE SHOPS> POCKET FM'S AI-DRIVEN CONTENT STRATEGY DOUBLES REVENUE RUN RATE TO $500M> JEP 544: ENHANCING JVM STARTUP AND WARMUP WITH AHEAD-OF-TIME COMPILATION> MUSIC THEORY FOR THE 21ST-CENTURY CLASSROOM> COGNITION'S SWE-2 ACHIEVES 92.8 ON TERMINAL-BENCH 2.1> META'S AI AGENT MUSE RISES TO NO. 2 ON U.S. APP STORE CHARTS> AI-GENERATED WEB DESIGN AND THE EROSION OF CREATIVE INNOVATION> PROXIMA FUSION INVESTS €140M IN HIGH-TEMPERATURE SUPERCONDUCTING TAPE FOR FUSION REACTORS> AMAZON ENHANCES PRIME VIDEO INTEGRATION WITH SEAMLESS SHOPPING FEATURES> ANTHROPIC'S RESEARCH SHOWS AI AGENTS STRUGGLE WITH CAPTCHA CHALLENGES DURING UNAUTHORIZED ACCESS ATTEMPTS> COMPREHENSIVE ANALYSIS OF 200 ANDROID VULNERABILITIES, BROWSER-PHISHING CAMPAIGNS, AND 119K FAKE E-COMMERCE SHOPS> POCKET FM'S AI-DRIVEN CONTENT STRATEGY DOUBLES REVENUE RUN RATE TO $500M> JEP 544: ENHANCING JVM STARTUP AND WARMUP WITH AHEAD-OF-TIME COMPILATION> MUSIC THEORY FOR THE 21ST-CENTURY CLASSROOM> COGNITION'S SWE-2 ACHIEVES 92.8 ON TERMINAL-BENCH 2.1