Anthropic announced the launch of its first public version of the high-performance Mythos series, 'Claude Payble 5,' on June 10. The company stated that new safeguards have been implemented to block responses in high-risk areas such as cybersecurity and biology, enabling a broader public release.
Claude Payble 5 achieved a score of 80.3% on the SWE-Bench Pro software engineering benchmark and surpassed the 90% mark on the Hex analysis benchmark for the first time. Anthropic described this model as specialized for complex research pipelines, long-term software development workflows, and long-horizon agentic tasks.
The rollout will be phased. Until June 22, it will be available at no additional cost for Pro, Max, Team, and sheet-based enterprise plans, transitioning to a paid credit system starting June 23. The plan will later restore basic features under subscription plans. Pricing is set at $10 per million input tokens and $50 per million output tokens, double the rates of Opus 4.8.
The model will be accessible through the Anthropic platform and AWS, Google Cloud, and Microsoft Foundry. Mythos 5 will be offered in a form without safety filters, available only to a select few cybersecurity defense organizations and infrastructure providers, including the previously approved Project Glasswing.
U.S. Congress Unveils First Draft of Federal AI Regulation: 'Great American AI Act'
On June 4, Republican Jay Obernolte (California) and Democrat Lori Trahan (Massachusetts), members of the House Energy and Commerce Committee, released a 269-page draft of the 'Great American AI Act (GAAIA).' This bipartisan bill aims to create the first comprehensive federal AI governance framework in the United States.
The bill targets 'large frontier developers' with annual revenues exceeding $500 million, addressing four key areas: frontier AI governance, workforce impact tracking, cybersecurity, and research and international cooperation. Violations could incur civil penalties of up to $1 million per day. The legislation seeks to establish a federal standard, preempting state laws related to AI model development for three years. However, the House Democratic AI Committee expressed opposition immediately after the bill's release, raising uncertainty about its path to enactment.
Google CEO Admits Lagging in Agentic Coding Compared to Anthropic and OpenAI
Sundar Pichai, CEO of Google, acknowledged in a recent podcast interview that the company is currently lagging in the area of agentic coding, which includes tool usage, command execution, and long-term tasks. Google has opted for a competitive strategy by launching a developer subscription tier at $100 per month, emphasizing affordability. Anthropic maintains its lead in the agentic coding market with Claude Code, while Microsoft is also strengthening its market presence by transitioning to a token-based Copilot billing system.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.