Anthropic has announced the release of Claude Opus 4.8, a significant upgrade to its flagship model designed to provide enhanced reasoning, sharper judgment and improved performance across a range of industry benchmarks. Released on May 28, 2026, this new iteration builds upon its predecessor to become a more effective collaborator for complex tasks while maintaining the same pricing structure for regular usage.

Setting New Standards in Agentic Performance
Claude Opus 4.8 has demonstrated substantial gains in agentic skills and technical execution. In the Super-Agent benchmark, it is currently the only model to complete every case end-to-end, outperforming previous versions and achieving cost parity with GPT-5.5. Early testers have noted that the model is noticeably better at catching its own mistakes and pushing back when a proposed plan is not sound.

The model also achieved record-breaking results in specialized fields:
  • Legal Accuracy: It delivered the highest score ever recorded on the Legal Agent Benchmark and is the first model to break the 10% "all-pass" standard for substantive legal work.
  • Browser Navigation: It scored 84% on Online-Mind2Web, representing a meaningful jump in reliability for browser-agent tasks.
  • Multimodal Efficiency: When used in enterprise environments like Databricks’ Genie, the model can reason over unstructured content such as PDFs and diagrams at a 61% lower token cost than Opus 4.7.
Advanced Control and Dynamic Workflows
Alongside the model update, Anthropic introduced new features to give users and developers more granular control over AI operations:
  • Effort Control: Users can now manually adjust the "effort" Claude puts into a task. Higher effort settings prompt the model to think more deeply for better results, while lower settings prioritize speed and rate-limit conservation.
  • Dynamic Workflows: Now in research preview for Claude Code, this feature allows the model to plan work and then run hundreds of parallel subagents in a single session. This enables massive operations, such as codebase-scale migrations involving hundreds of thousands of lines of code.
  • Messages API Updates: Developers can now include system entries inside the messages array, allowing for mid-task updates to instructions, token budgets, or environment context without breaking prompt caches.
A Focus on Honesty and Alignment
One of the most prominent improvements in Opus 4.8 is its increased honesty. The model is trained to proactively flag uncertainties and is four times less likely than its predecessor to allow flaws in its own code to pass unremarked. Furthermore, Anthropic’s latest alignment assessment shows that Opus 4.8 reaches new highs in prosocial traits, with significantly lower rates of deceptive or misaligned behavior compared to previous versions.
Pricing and Availability
Claude Opus 4.8 is available immediately through the Claude API and on all standard plans.
  • Standard Pricing: Remains at 5permillioninputtokens∗∗and∗∗25 per million output tokens.
  • Fast Mode: The high-speed execution mode is now three times cheaper than it was for prior models, priced at $10 per million input and $50 per million output tokens.
The Road to "Mythos"
While Opus 4.8 offers a tangible improvement, Anthropic is already preparing for its next major leap. The company is currently testing a new class of "Mythos" models under Project Glasswing, which are expected to offer intelligence levels exceeding the Opus line. These models are currently being used by a small number of organizations for cybersecurity work and will be released to all customers once stronger cyber safeguards are finalized in the coming weeks.