Key takeaways

• Claude Opus 4 (2025-05-14) is a proprietary, Transformer-based model from Anthropic, designed for advanced text generation, complex multi-step tasks, and intricate coding.
• It excels in sustained reasoning and delivers strong performance on long-running tasks requiring thousands of steps, such as scientific reasoning, achieving 99.69% accuracy on easy-level science questions with low toxicity.
• A significant limitation is its verbosity, leading to 0% accuracy on tasks requiring concise, precise answers and nuanced constraint satisfaction, particularly evident in medical queries (e.g., Smart 07-07 dataset).
• The model's tendency to provide overly verbose and detailed responses, even when brief answers are expected, results in frequent mismatches with ground truth in precision-focused tasks, failing to generate concise and accurate responses in clinical contexts.
• Despite its comprehensive understanding of complex topics like the Epstein-Barr Virus, its inability to be direct and concise leads to poor performance on benchmarks requiring exact, brief matches to ground truth.

Technical specifications

Context length

200k tokens

Max output

—

Architecture

Transformer

Modalities

text+image->text

Parameters

~1000B

Open weights

No

Licence

Proprietary

Regions

USA

Vendor

Anthropic

Provider model key

anthropic/claude-opus-4-20250514