Anthropic says Opus 4.6 has a 1M-token context window in beta, scored 90.2% on BigLaw Bench, the highest of any Claude model, and boosts agentic capabilities
David Gewirtz /ZDNET:
Context & Ripple Effects
Anthropic’s Opus line was already positioned around coding, agents and computer use with the prior Opus 4.5 release. Opus 4.6 adds a beta 1M-token context window and a higher reported BigLaw Bench result to that product arc.
The model upgrade arrives alongside a research preview of parallel agent teams in Claude Code, making context capacity and agent coordination mutually relevant parts of Claude’s workflow pitch.
First-order effects
- Claude users can test a beta 1M-token context window in Opus 4.6, potentially keeping larger bodies of material in a single model interaction.
- Anthropic can market Opus 4.6’s reported 90.2% BigLaw Bench score as its strongest Claude-model result, while tying the release to improved agentic capability.
Second-order effects
- Long-context and agent-oriented model offerings become a more direct comparison point for competing AI providers, especially for users evaluating complex document and coding workflows.
- Teams adopting Claude agents may place greater value on how well models retain and prioritize supplied material; Anthropic’s own claim of deeper focus on difficult task components becomes part of that evaluation alongside benchmark scores.
Third-order effects
- If larger context windows and multi-agent tooling mature together, AI products may increasingly compete as work surfaces for persistent, multi-step tasks rather than as standalone chat interfaces.
- The differentiator could shift from raw context-window size toward context engineering: which information agents retain, prioritize and use reliably across complex workflows.
The trend: This is one data point in the shift toward agentic AI systems that combine larger working context with coordinated task execution.