Anthropic has launched an upgraded version of its primary AI model, Opus 4.7, just one week after restricting the release of its most advanced system, Mythos. Concurrently, the company is pursuing a new round of funding that could potentially double its valuation to $800 billion or higher.
The company stated on Thursday that Opus 4.7 demonstrates improved performance in software engineering, alongside enhancements in instruction-following capabilities and image recognition. However, Anthropic explicitly noted that the model's overall capabilities in areas like cybersecurity are inferior to those of Mythos. The company even conducted experimental "differential reduction" of its cyber attack and defense abilities during the training process.
This deployment strategy reflects Anthropic's careful balancing act between model capability and safety boundaries, while also paving the way for the eventual broad commercialization of Mythos-level models.
**Opus 4.7: Enhanced Programming Skills, Reduced Cyber Capabilities**
Anthropic indicated that the core improvements in Opus 4.7 focus on the software engineering domain, including handling complex programming tasks that previously required more human supervision. Additionally, the model's image recognition capabilities have been enhanced, enabling it to identify detailed information within complex charts or images.
Regarding safety mechanisms, Anthropic has built an automatic detection and interception system into Opus 4.7, which can identify and block requests involving prohibited or high-risk cybersecurity applications. The company stated, "We will gather experience from the practical deployment of these protective measures to advance the goal of eventually achieving a widespread release for Mythos-level models."
Notably, Anthropic proactively intervened during the training phase of Opus 4.7, specifically implementing a "differential reduction" of its cyber attack and defense capabilities—a practice relatively uncommon in large language model development.
**Concerns Over Mythos Capabilities Lead to Restricted Release**
Last week, Anthropic disclosed that its Mythos system possesses highly dangerous cyber attack capabilities—under user direction, the model can identify and exploit security vulnerabilities in "every major operating system and every major browser." Due to safety concerns, Anthropic decided to make Mythos available only to a select group of enterprise clients, specifically to help them strengthen their own software systems, rather than releasing it to the general public.
Against this backdrop, the release of Opus 4.7 fills the market gap left by Mythos's lack of broad commercial availability, while also providing Anthropic with an opportunity to validate its safety mechanisms in real-world deployment environments.
**Competitive Landscape and Funding Prospects**
Anthropic is currently engaged in fierce competition with OpenAI, as both companies strive to deploy more advanced AI models and attract more enterprise customers to adopt their paid services. In recent months, Anthropic's AI programming products have gained significant momentum, and consumer interest continues to grow, despite disagreements with the Pentagon over AI safety standards.
On the funding front, Anthropic's most recent valuation was $380 billion. It is reported that the company is currently soliciting offers from investors for a new funding round, with a potential valuation reaching approximately $800 billion or higher—more than double the previous round's valuation.