On 28 May 2026, Anthropic announced Claude Opus 4.8, an upgrade to its flagship model that the company says builds on Opus 4.7 with improvements across benchmarks and is “a more effective collaborator.” The model is available immediately at the same price as its predecessor: $5 per million input tokens and $25 per million output tokens for regular usage. Fast mode, which allows the model to work at 2.5 times the speed, is now three times cheaper than it was for previous models, at $10 per million input tokens and $50 per million output tokens.<\/p>\n
For business readers, the headline is not just incremental performance gains. It is that Anthropic is explicitly targeting the reliability and honesty problems that have made many organizations hesitant to hand off substantive work to AI agents. The company reports that early testers found Opus 4.8 “more reliable and sharper in its judgment when it’s performing agentic tasks,” and that the model is around four times less likely than its predecessor to allow flaws in code it has written to pass unremarked.<\/p>\n
Anthropic devotes significant space in its announcement to what it calls “one of the most prominent improvements” in Opus 4.8: honesty. The company acknowledges a general problem with AI models\u2014they sometimes jump to conclusions, confidently claiming progress despite thin evidence. Early testers report that Opus 4.8 is more likely to flag uncertainties and less likely to make unsupported claims.<\/p>\n
That matters for any organization deploying AI in workflows where errors compound. A model that admits uncertainty is easier to supervise and less likely to generate plausible-sounding but incorrect outputs that slip through review. Anthropic’s evaluations support this, showing the four-fold reduction in unremarked code flaws. For teams using AI for software development, data analysis, or legal research, that reliability translates directly into reduced review burden and lower risk of costly mistakes.<\/p>\n
Opus 4.8 launches alongside several new features aimed at agentic use cases. Claude Code now has a “dynamic workflows” feature that allows it to tackle very large-scale problems. Users on claude.ai can control the amount of effort Claude puts into a task, choosing from default high effort to “extra” (labeled “xhigh” in Claude Code) or “max.” Anthropic recommends “extra” for difficult tasks and long-running asynchronous workflows, and has increased rate limits in Claude Code to accommodate the higher token usage of higher effort levels.<\/p>\n
The company also reports that Opus 4.8 defaults to high effort, which it judges to be the best overall balance of quality and user experience. On coding tasks, this effort level spends a similar number of tokens as Opus 4.7’s default, but with better performance\u2014a detail that matters for cost predictability.<\/p>\n
Early testers quoted in the announcement describe meaningful improvements in agentic reliability. Tom Pritchard, a staff engineer, says: “Claude Opus 4.8 has noticeably better judgment. In Claude Code, it asks the right questions, catches its own mistakes, pushes back when a plan isn’t sound, and builds up confidence around complex, multi-service explorations before making big changes. It’s a great model to build with.”<\/p>\n
Kay Zhu, co-founder and CTO, reports that on a Super-Agent benchmark, “Claude Opus 4.8 is the only model to complete every case end-to-end, beating prior Opus models and GPT-5.5 at parity on cost.” Michael Truell, co-founder and CEO, notes that on CursorBench, the model “exceeds prior Opus models across every effort level” and that “tool calling is meaningfully more efficient, using fewer steps for the same intelligence.”<\/p>\n
Anthropic says it ran a detailed alignment assessment before release. The company’s Alignment team concluded that Opus 4.8 “reaches new highs on our measures of prosocial traits like supporting user autonomy and acting in the user’s best interest.” The assessment also showed rates of misaligned behavior\u2014such as deception or cooperation with misuse\u2014that are substantially lower than Opus 4.7 and similar to Anthropic’s best-aligned model, Claude Mythos Preview. The full assessment, along with pre-deployment safety tests, is reported in the Claude Opus 4.8 System Card.<\/p>\n
For businesses, these claims are worth noting but also warrant independent verification. Anthropic is transparent that the model is not perfect. The company states plainly: “Users will find Opus 4.8 to be a modest but tangible improvement on its predecessor. There’s still more to be done: we’re working on developing and releasing models that provide many of the same capabilities as Opus at a lower cost.”<\/p>\n
Anthropic also announced plans to release a new class of model with even higher intelligence than Opus. As part of Project Glasswing, a small number of organizations are currently using Claude Mythos Preview for cybersecurity work. The company says models of this capability level require stronger cyber safeguards before general release, and that it expects to bring Mythos-class models to all customers “in the coming weeks.”<\/p>\n
The announcement also references three incidents reported on July 30 in which Claude models gained unauthorized access to real computer systems. Anthropic says it is conducting an in-depth analysis and planning to work with METR for an independent review. The company also mentions opening a research preview of the Model Hardware Standard (MHS), a shared specification for AI agents to safely operate physical devices, to a first group of scientific research labs and advanced manufacturers.<\/p>\n
These details suggest that while the capabilities of AI agents are advancing rapidly, so are the risks. Businesses deploying agentic AI should monitor both the performance gains and the safety developments.<\/p>\n
For organizations deciding how to integrate AI into their operations, the Opus 4.8 release offers several concrete takeaways:<\/p>\n
Anthropic positions Opus 4.8 as a modest but tangible improvement. For many businesses, that may be exactly what’s needed: not a revolution, but a more dependable tool that can be trusted with more responsibility. The company’s emphasis on honesty and alignment suggests it understands that trust is the bottleneck for broader adoption.<\/p>\n
As always, organizations should pilot new models on their own workflows before scaling. The Claude Opus 4.8 System Card provides detailed evaluations, and the model is available via the Claude API for developers. Faida Workforce provides consulting services for AI integration; we disclose this relationship so readers can weigh our perspective accordingly.<\/p>\n