Anthropic introduces Opus 4.5, cuts API pricing, and enables much longer Claude chats (arstechnica.com)

Anthropic released Opus 4.5 and rolled out app- and API-level changes designed to make Claude more robust for long, interactive sessions and more cost-effective to use. In consumer apps (web, mobile, desktop) Claude will no longer abruptly end conversations when a hard 200k-token context window is reached; instead it now performs behind-the-scenes summarization and context compaction to keep important details while discarding extraneous content. Anthropic says the same principles are available to developers via context management tools in the API, and it also reduced API pricing to encourage broader adoption. Technically, Opus 4.5 is a notable step forward on coding and agentic tasks: it scored 80.9% on SWE-Bench Verified, edging out OpenAI’s GPT-5.1-Codex-Max (77.9%) and Google’s Gemini 3 Pro (76.2%). The model shines on agentic coding and tool-use benchmarks but still trails GPT-5.1 on visual reasoning (MMMU). For the AI/ML community this means a more competitive frontier model for building long-lived, stateful assistants and agent pipelines, plus practical primitives (summarization/context compaction) to manage very long contexts without user-visible failures.