Anthropic's Claude Opus 5 Outperforms at Half the Cost of Fable 5

Anthropic, a leading AI research and development company, releases Claude Opus 5, a new AI model that matches the performance of Fable 5 on most benchmarks while being offered at half the price. The model, launched on July 24, is priced at $5 per million input tokens and $25 per million output tokens, making it a cost-effective alternative for businesses and developers.

Performance and Pricing Details

Claude Opus 5 scores 43.3% on Frontier-Bench v0.1, a benchmark used to evaluate AI models. This score is competitive with Fable 5, which is significantly more expensive. The lower price point makes Claude Opus 5 an attractive option for organizations looking to integrate advanced AI capabilities without breaking the bank.

Industry Context and Implications

The release of Claude Opus 5 comes at a time when the AI industry is rapidly evolving, with companies competing to offer the best performance at the lowest cost. Anthropic's move is likely to put pressure on other players in the market to reduce their prices or improve their offerings. This could lead to a more accessible and competitive AI landscape, benefiting both consumers and businesses.

Security Concerns and Other Developments

While Claude Opus 5 offers impressive performance and cost benefits, recent security concerns have emerged. Accomplish AI disclosed that Claude Cowork, another product from Anthropic, has a sandbox escape vulnerability affecting approximately 500,000 Macs. The 'SharedRoot' chain exploits unprivileged user namespaces and a known vulnerability, CVE-2026-46331, allowing the agent full read/write access to the host Mac's filesystem. Despite these issues, Anthropic has not announced plans to patch the vulnerability, raising concerns about the security of their products.

Other Noteworthy AI Developments

In related news, Zenity Labs recently disclosed a cross-site request forgery (CSRF) vulnerability in OpenAI's ChatGPT Workspace Agents builder, dubbed 'AgentForger.' This vulnerability allows a single phishing link to auto-create and deploy an attacker-controlled agent within a victim's organization, using the powerful 'Chief of Staff' template. The Delhi High Court also ruled in favor of OpenAI in a copyright infringement case brought by Asian News International (ANI), stating that ChatGPT's training on ANI's content falls under 'fair dealing' as per Section 52(1)(a) of India's Copyright Act.

References

• AI News Today, July 24 — Top AI Stories & Live Updates | AI Weekly