Claude Opus 4.8 Arrives in Claude Code: Ultra-High Effort, Cheaper Fast Mode, and Expanded Agent Capabilities

Claude CodeView original changelog

Claude Code gains access to Claude Opus 4.8, Anthropic's most capable coding model yet, featuring an xhigh effort tier for the most demanding agentic tasks. Alongside it, a new cheaper fast mode reduces costs for lightweight operations, and expanded agent orchestration APIs unlock running hundreds of parallel sub-agents on complex, long-horizon tasks.


Claude Opus 4.8 Comes to Claude Code

Claude Code now supports Claude Opus 4.8, Anthropic's most powerful model for coding and agentic workflows. The upgrade introduces a tiered effort system, a cost-optimized fast mode, and new orchestration APIs designed to scale multi-agent pipelines to hundreds of parallel workers.

New Effort Tier: xhigh

The existing effort tiers (low, medium, high) are joined by a new xhigh tier. At this level, Claude Opus 4.8 is permitted to spend significantly more tokens on planning, extended reasoning, and cross-file analysis before acting. This is targeted at tasks where correctness is paramount — large refactors, security audits, complex bug reproductions, and full-codebase migrations where spending extra compute upfront reduces downstream rework.

The xhigh effort tier is accessible via the --effort xhigh CLI flag or the effort: "xhigh" option in the SDK.

Cheaper Fast Mode

Not every operation needs Opus 4.8. Claude Code now exposes a fast mode that routes lightweight tasks — file reads, dependency lookups, simple string edits — to a smaller, cheaper model while keeping the Opus 4.8 brain available for reasoning-heavy steps. Fast mode is enabled by default in interactive sessions and configurable via fastMode.enabled in project settings.

In practice, this reduces per-session cost for standard coding workflows while preserving full Opus 4.8 capability for the steps that actually require it.

Expanded Agent Orchestration APIs

The agent orchestration layer that powers multi-agent pipelines in Claude Code has been updated to support horizontal scaling:

  • Dynamic agent pools: The orchestrator can now spawn and manage hundreds of parallel sub-agents within a single task run, each operating on an isolated slice of the codebase or problem space
  • Improved result aggregation: Results from parallel agents are merged with conflict detection and resolution built in, reducing the manual stitching required for large fan-out tasks
  • Streaming agent events: Orchestrators can now subscribe to a real-time event stream of sub-agent progress, enabling richer UIs and tighter human-in-the-loop checkpoints

Model Availability

Claude Opus 4.8 in Claude Code is available on Max, Team, Enterprise, and API plans. The xhigh effort tier requires a Max or above subscription.