tech
Anthropic introduces cheaper, more powerful, more efficient Opus 4.5 model
Longer chats address a long-standing criticism of Claude.
TL;DR
- Opus 4.5 improves conversation memory, preventing abrupt stops by summarizing earlier parts of long chats.
- The model achieved 80.9% accuracy on the SWE-Bench Verified benchmark, surpassing OpenAI's GPT-5.1-Codex-Max and Google's Gemini 3 Pro.
- Opus 4.5 shows improved efficiency, using significantly fewer tokens for comparable or better performance.
- New developer features include an 'effort' parameter for balancing efficacy and token usage.
- Claude Code is now available in the desktop Claude apps.
- API pricing for Opus 4.5 has been reduced to $5 (input)/$25 (output) per million tokens.