Anthropic has introduced Claude Opus 4.7, following the recent release of Opus 4.6. This new version is available to Pro, Max, Team, and Enterprise subscribers, along with access through APIs, Amazon Bedrock, Vertex AI, and Microsoft Foundry. The pricing structure remains the same, costing $5 per million tokens for input and $25 for output.
Impressive Advances in Agentic Coding
Opus 4.7 showcases remarkable improvements in agentic coding benchmarks. According to SWE-bench Pro, it achieves a score of 64.3%, surpassing its predecessor Opus 4.6 at 53.4%, GPT-5.4 at 57.7%, and Gemini 3.1 Pro at 54.2%. Furthermore, in SWE-bench Verified, Opus 4.7 reaches an impressive 87.6%.
Exceptional Reasoning Ability
The model excels in doctoral-level reasoning, scoring 94.2% in GPQA Diamond assessments, which is comparable to GPT-5.4 Pro’s score of 94.4%. Opus 4.7 is rigorously benchmarked across 13 evaluations against Opus 4.6, GPT-5.4, Gemini 3.1 Pro, and Mythos Preview.
Enhancements in Visual Processing
One significant upgrade in Opus 4.7 is its ability to process images at three times the resolution of Opus 4.6. This enhancement greatly improves the quality of generated interfaces, slides, and documents.
New API Functionality
Developers will also benefit from a new effort tier labeled “xhigh,” positioned between “high” and “max.” This allows for a more nuanced balance between reasoning depth and response latency. Performance in agentic coding varies with effort levels (low, medium, high, xhigh, max), and Opus 4.7 consistently outperforms Opus 4.6 across all tiers, delivering higher efficiency with lower token budgets.
A Model Designed for Extended Autonomy
Anthropic aims for Opus 4.7 to tackle long, complex tasks with minimal supervision. The model self-verifies its outputs, maintains coherence across its extensive context window of one million tokens, and can manage multi-session projects comprehensively.
Strategy Overview
It’s important to note that Opus 4.7, while advanced, is not the most powerful model from Anthropic. Claude Mythos Preview is reserved for select cybersecurity partners as part of Project Glasswing and currently achieves a score of 77.8% on SWE-bench Pro.
In summary, Anthropic is adopting a dual strategy, deploying Opus for commercial use while keeping Mythos at the forefront of cybersecurity applications.
