What's new in Claude Opus 4.7
6 hours ago
- #Agentic AI
- #Claude Opus 4.7
- #AI Models
- Claude Opus 4.7 is the most capable generally available model with high autonomy and excels at long-horizon agentic work, knowledge work, vision tasks, and memory.
- It introduces high-resolution image support up to 2576px/3.75MP, enhancing vision-heavy workloads like document understanding, with pixel-accurate coordinates and no scale-factor math required.
- The model features a new effort parameter to tune intelligence vs. token spend, with 'xhigh' recommended for coding and agentic use cases, and 'high' for intelligence-sensitive tasks.
- Task budgets allow setting an advisory token limit for full agentic loops, helping the model prioritize work, with a minimum of 20k tokens, distinct from the hard max_tokens cap.
- Breaking changes include removal of extended thinking budgets (adaptive thinking only), non-default temperature/top_p/top_k parameters causing errors, and thinking content omitted by default unless opted in.
- A new tokenizer may use 1x to 1.35x more tokens, affecting token counts, and prompts may need updates for max_tokens and cost control, with improvements in knowledge-worker tasks and memory use.