San Francisco, CA — Anthropic has released Claude Opus 4.7, its most intelligent Opus model to date, now available on Amazon Bedrock. The new model promises significant improvements in agentic coding, knowledge work, and long-running task execution, underpinned by Bedrock’s next-generation inference engine.
“Opus 4.7 is designed for the most demanding production workloads,” said an Anthropic spokesperson. “It reasons through ambiguity, self-verifies its output, and stays on track over its full 1 million token context window.”
Background
Anthropic’s Opus series has been a flagship for advanced reasoning and code generation. The predecessor, Claude Opus 4.6, set benchmarks in agentic coding. Opus 4.7 extends that lead with stronger long-horizon autonomy and systems engineering.
The model is hosted on Amazon Bedrock, which now features a new inference engine with dynamic scheduling and scaling logic. This engine improves availability for steady-state workloads while accommodating rapid scaling services. It also provides zero operator access, ensuring customer prompts and responses remain private from both Anthropic and AWS operators.
Key Improvements
- Agentic coding: Scores 64.3% on SWE-bench Pro, 87.6% on SWE-bench Verified, and 69.4% on Terminal-Bench 2.0.
- Knowledge work: Reaches 64.4% on Finance Agent v1.1, with improved document creation and multi-step research.
- Long-running tasks: Maintains performance over its full 1M token context, with better ambiguity handling and self-verification.
- Vision: Adds high-resolution image support for charts, dense documents, and screen UIs.
“Opus 4.7 handles underspecified requests by making sensible assumptions and stating them clearly,” the spokesperson added. “It then verifies its own work to improve quality from the start.”
What This Means
For developers and enterprises, Claude Opus 4.7 on Bedrock reduces the gap between AI capabilities and production requirements. The improvements in agentic coding mean teams can deploy more autonomous agents for complex systems engineering.
The zero-operator access and new inference engine also address security and scalability concerns, making it viable for sensitive data workloads. However, Anthropic notes that users may need to adjust prompting and harness tweaks to fully leverage the model.
“This is not just a minor upgrade,” said an AWS AI strategist. “The dynamic scaling logic in Bedrock’s new engine directly addresses the pain point of inference availability during demand spikes.”
Getting Started
Users can access Claude Opus 4.7 via the Amazon Bedrock console, selecting it under the Playground menu. It is also available programmatically through the Anthropic Messages API, Bedrock runtime, or the Converse API.
An example prompt from Anthropic: “Design a distributed architecture on AWS in Python that should support 100k requests per second across multiple geographic regions.” The model can now reason through such open-ended tasks more thoroughly.
For further details, see Anthropic’s prompting guide.