Anthropic Launches Claude Opus 4.8 With Major Coding, Reasoning, and Agentic Gains
Anthropic has officially released Claude Opus 4.8, a major upgrade to its flagship model designed to handle complex tasks, agentic coding, and high-autonomy professional workflows. The new version launches at the same price point as its predecessor and introduces novel features including effort level control on claude.ai, dynamic workflows in Claude Code, and a three-times cheaper “fast mode” tier for high-speed execution.
Context
Claude Opus 4.8 represents a direct evolution of the Opus 4.7 model released in April 2026. Anthropic continues its rapid pace of updates to its flagship commercial offerings. While experimental models like Claude Mythos remain restricted due to safety precautions, the company focused on enhancing practical reliability and safety alignment for its public flagship, successfully mitigating previous issues such as comment verbosity and inefficient tool calling.
Details
Benchmarks and Performance Highlights
Third-party tests and Anthropic’s evaluations show substantial progress across key agentic capabilities:
- CursorBench: The new model exceeded prior Opus versions across every effort level. Tool calling was notably more efficient, executing tasks in fewer steps without losing accuracy.
- Legal Agent Benchmark (CoCounsel): Opus 4.8 delivered the highest score ever recorded on this benchmark, becoming the first LLM to break 10% on the strict “all-pass” standard for substantive legal work.
- Online-Mind2Web: Scoring 84% on web-agent and browser tasks, the model demonstrated a significant jump over both Opus 4.7 and GPT-5.5.
- Super-Agent Benchmark: Claude Opus 4.8 was the only model to successfully complete every agent scenario end-to-end while remaining cost-competitive with alternative options.
New Tooling and Developer Features
Alongside the core model, Anthropic announced several productivity upgrades:
- Effort Controls on claude.ai: Users now have granular control over the depth of thinking and computational effort Claude applies to solving a given prompt.
- Dynamic Workflows in Claude Code: A developer-focused feature allowing Claude to spawn and coordinate hundreds of parallel subagents to solve complex, multi-service codebases.
- Optimized Fast Mode: The Opus 4.8 fast mode, capable of running up to 2.5× faster, is now 3× cheaper compared to previous generations.
Early feedback from beta testers has been highly positive. Scott Wu, CEO of Cognition (creators of Devin), noted that Opus 4.8’s cleaner tool calls and instruction following translate directly to faster capability gains for autonomous engineering workloads. Michael Ran, Senior Investment Associate, praised the model’s reliability in long-running analyses, highlighting its proactive ability to flag errors and inconsistencies in source data, significantly reducing noise in high-stakes workflows.