Anthropic has released Claude Sonnet 5 with an always-on 1 million token context window, eliminating the need for users to opt into larger contexts or use workarounds. However, the author argues that this change does not end the need for context discipline, but rather shifts its focus from managing limited windows to optimizing within the expanded capacity. The new tokenizer and degradation data suggest that efficient use of the large context is still crucial for performance. AI
IMPACT Shifts focus from context rationing to efficient use of expanded context windows, potentially impacting how users interact with LLMs.
RANK_REASON Frontier-lab model release with system card [lever_c_demoted from frontier_release: ic=2 ai=1.0]
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →