Anthropic Unveils Opus 4.6: A Smarter Coding AI with a Million-Token Leap, But User Experience Takes a Hit

Anthropic has officially launched Opus 4.6, positioning it as the “smartest AI coding model ever made.” This iteration significantly enhances its predecessor’s coding capabilities, demonstrating improved planning, sustained agentic task execution, and more reliable operation within large codebases. A notable new feature is the beta release of a 1-million token context window, a first for Opus-class models, aimed at reducing the need for context compaction. Opus 4.6 has achieved impressive benchmark results, including an “absurd score” on Humanity’s Last Exam V2, pushing into the 70% range and surpassing previous high scores on a benchmark designed to challenge AI agents. The model also introduces an experimental orchestration layer supporting “swarms and teams” of parallel agents, exemplified by the successful, albeit costly ($20,000 API spend), construction of a Rust-based C compiler capable of building the Linux kernel from scratch.

Despite its enhanced intelligence, early user feedback on Opus 4.6 indicates a nuanced experience. While capable of solving problems Opus 4.5 could not, some users report a perceived decrease in interaction speed and a regression in “prose,” making conversations feel more templated and less varied. The pricing for Opus 4.6 remains consistent at $5/million input tokens and $25/million output tokens, which is significantly higher—two to four times—than comparable OpenAI models like GPT-5.2. Utilizing the new 1-million token context window further escalates costs, doubling input and nearly output token prices. On the API front, Anthropic has implemented a change to block “partial turn prefill” to mitigate misuse, preventing developers from seeding incomplete responses. This move highlights ongoing industry tensions around model steering, data privacy (reasoning tokens), and the potential for provider lock-in as models become more deeply integrated with proprietary server-side histories. The release also sparks speculation that Opus 4.6 might be a re-branded Sonnet 5, given its new capabilities and premium pricing strategy.