Synopsis
Anthropic on Thursday released Claude Opus 4.7, the latest upgrade in its flagship Opus AI model family. Users will now be able to delegate work that previously required close supervision to the AI assistant. The model demonstrates stronger instruction-following, higher consistency over extended workflows, and the ability to self-verify outputs before delivering results.Listen to this article in summarized format
The company said the new model delivers improved performance in advanced software engineering and long-duration task execution.
What changes with Opus 4.7?
Users will now be able to delegate work that previously required close supervision to the AI assistant. The model demonstrates stronger instruction-following, higher consistency over extended workflows, and the ability to self-verify outputs before delivering results.
The update also brings enhanced multimodal capabilities. Opus 4.7 can process higher-resolution images, where the longer dimension can be capped at 2,576 pixels, enabling more accurate interpretation of dense visuals such as diagrams and screenshots. Anthropic said the model also produces more polished, professional outputs, including interfaces, documents, and presentations.
On safety, Anthropic said Opus 4.7 maintains a similar profile to its predecessor, with lower rates of misaligned behaviour and improved resistance to prompt injection attacks, though some weaknesses persist.
Anthropic also introduced new features, including finer control over reasoning effort and tools to manage token usage, alongside updates to its developer platform and coding workflows.
How does it compare with other models from rivals?
Per various benchmarks, Opus 4.7 remains ahead of its predecessor Opus 4.6, OpenAI’s GPT-5.4, and Google’s Gemini 3.1 Pro, particularly in agentic coding, terminal coding, and financial analysis.
However, it lags GPT-5.4 and Gemini 3.1 Pro in agentic search and multidisciplinary reasoning with tools. It also slightly trails Opus 4.6 in cybersecurity vulnerability reproduction.
How does it compare with Claude Mythos Preview?
Despite these upgrades, Opus 4.7 is positioned below the company’s most advanced system, Claude Mythos Preview, with the latter outperforming the latest Opus model in 10 of 11 benchmarks, except agentic search.
The release follows the Anthropic-led startup’s Project Glasswing, announced earlier this month, bringing together major tech firms to secure critical software systems. It comes as the company claims that AI models have reached coding capabilities that can surpass most humans in identifying and exploiting vulnerabilities.
As part of that strategy, Opus 4.7 is the first model to deploy new safeguards aimed at detecting and blocking high-risk cybersecurity misuse.
The company said it has intentionally limited the model’s cyber capabilities relative to Mythos Preview and will use real-world deployment data from Opus 4.7 to refine protections ahead of broader releases.
When will the model be available and at what pricing?
The model is available across Claude products, APIs, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry starting today.
Pricing remains unchanged, the same as Opus 4.6, at $5 per million input tokens and $25 per million output tokens.
