Opus 4.7 Part 2: Capabilities and Reactions

2023-08-29 · Source: Don't Worry About the Vase · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Data Science & Analytics · Depth: Advanced, extended

Summary

Anthropic's Claude Opus 4.7 is presented as a significant advancement over its predecessor, Claude Opus 4.6, particularly in advanced software engineering and handling complex, long-running tasks with rigor and consistency. It features improved vision with higher image resolution and enhanced creativity for professional tasks. While generally considered the most intelligent model in its class for many purposes, it exhibits "jaggedness," including strange refusals, verbosity, and issues with adaptive thinking. The model is available across Claude products, API, Amazon Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry, maintaining the same pricing as Opus 4.6 at $5 per million input tokens and $25 per million output tokens. Benchmarks show mixed results, with gains in areas like visual reasoning and economically valuable tasks (GDPVal-AA), but regressions in others such as OpenAI MRCR v2 and OSWorld, often attributed to changes in its adaptive thinking implementation.

Key takeaway

For AI engineers and CTOs evaluating large language models, Claude Opus 4.7 represents a powerful, albeit idiosyncratic, tool. Its superior coding and agentic capabilities make it a strong contender for demanding development tasks, but its "non-sycophantic" nature and adaptive thinking quirks necessitate a shift in prompting strategy. Be prepared to "treat the model well" and justify tasks, or consider retaining Opus 4.6 for use cases requiring more predictable instruction following or less complex reasoning.

Key insights

Claude Opus 4.7 offers enhanced intelligence and autonomy but requires careful handling due to its distinct behavioral quirks.

Principles

• Model performance varies significantly across benchmarks and use cases.
• Effective interaction requires treating the model like a capable coworker.
• Adaptive thinking can lead to inconsistent performance in non-coding tasks.

Method

To optimize Claude Opus 4.7, specify tasks upfront, minimize user interactions, use auto mode, and consider adjusting custom instructions to align with its less sycophantic and more autonomous nature.

In practice

• Use Claude Opus 4.7 for complex coding and long-running agentic workflows.
• Retain GPT-5.4 for web searches and factual checks.
• Experiment with system_prompt="." for potentially better results.

Topics

• Claude Opus 4.7
• AI Model Benchmarking
• Adaptive Thinking
• Software Engineering Performance
• Multimodal AI

Best for: AI Engineer, CTO, VP of Engineering/Data, AI Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• Claude's Opus 4.7 Review — Claude Opus 4.7 significantly improves agentic coding, instruction following, and multimodal understanding.
• My Honest Review of Claude Opus 4.6: Is It Worth the Hype? — Claude Opus 4.6 offers superior coding and reasoning capabilities with adaptive thinking and extended context windows.
• Claude Opus 4.8: Capabilities and Reactions — Comprehensive AI model evaluation requires diverse benchmarks, user feedback calibration, and understanding model "welfare" for accurate assessment.
• Claude Opus 4.7: Most Powerful Coding Model Ever! Beats EVERYTHING! (Fully Tested) — Claude Opus 4.7 significantly advances coding, web development, and vision, but at a higher token cost.
• Introducing Claude Opus 4.7 — Claude Opus 4.7 offers significant advancements in software engineering, multimodal understanding, and long-running task consistency with enhanced safety.
• Anthropic released Claude Opus 4.6, with major gains across coding, knowledge, and reasoning benchmarks. — AI models are rapidly advancing in capabilities and context, driving massive infrastructure investment and strategic corporate consolidation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Don't Worry About the Vase.