Anthropic releases a new Opus model amid Mythos Preview buzz
New 'generally available' model improves software engineering but scores lower than private cybersecurity-focused Mythos on every test.
Anthropic has launched Claude Opus 4.7, positioning it as the most powerful model in its generally available lineup. The new iteration shows marked improvements over Opus 4.6, particularly for advanced software engineering tasks that previously required significant manual guidance. It also delivers enhanced capabilities in analyzing images, following complex instructions, and generating creative content for slides and documents. Early testing involved major customers like Intuit, Replit, Notion, and Databricks, with pricing holding steady at $5 per million input tokens and $25 per million output tokens.
Despite these advancements, Opus 4.7 represents a strategic step back in raw capability. In its system card, Anthropic explicitly states the model does not advance the company's 'capability frontier,' as the recently announced Claude Mythos Preview outperformed it on every relevant evaluation. For security reasons, Mythos Preview remains privately available only to select partners like Nvidia, JPMorgan Chase, and major tech firms. Anthropic confirmed it experimentally reduced Opus 4.7's cyber capabilities during training as part of a new safety protocol, deploying it with additional safeguards. The findings from this rollout are intended to pave the way for a future broad release of 'Mythos-class' models. Security professionals can apply to test Opus 4.7 for vulnerability research with fewer restrictions through Anthropic's new Cyber Verification Program.
- Claude Opus 4.7 is Anthropic's new top-tier generally available model, showing improved performance in complex coding and creative tasks over Opus 4.6.
- The model is explicitly less capable than the private Claude Mythos Preview, scoring lower on every evaluation as Anthropic deliberately limited its advanced cyber functions for safety testing.
- Priced at $5/$25 per million tokens, it's available now with enhanced safeguards, while Mythos Preview remains restricted to partners like Nvidia and JPMorgan.
Why It Matters
This release highlights the industry's cautious, safety-first approach to deploying cutting-edge AI, prioritizing controlled testing over raw capability in public models.