Anthropic released Claude Opus 4.8 on May 28, 2026, bringing meaningful improvements to its flagship model without changing the price. At $5 per million input tokens and $25 per million output tokens, the update lands as a straightforward upgrade for developers and enterprises already building on Opus 4.7.
The most discussed improvement is a sharper internal error-detection system. Claude Opus 4.8 is approximately four times less likely to let code flaws pass without flagging them, compared to its predecessor. For teams running autonomous coding agents or automated report generation, this matters more than benchmark scores: an AI that quietly delivers broken output is far more dangerous than one that fails loudly. The update shifts the failure mode from silent to visible.
On the performance side, Claude Opus 4.8 scores 84% on the Online-Mind2Web benchmark, which tests browser-based agent tasks, and is the first model to exceed 10% on the Legal Agent Benchmark’s all-pass standard, a strict test of legal reasoning accuracy. Across coding, reasoning, and agentic tasks, it outperforms Opus 4.7.
Three new features are worth noting for teams deploying Claude at scale. The first is Effort Control, available on Claude.ai and in Cowork integrations. Users can now select the level of thinking effort applied to a task: higher settings prompt Claude to reason more deeply before answering. This is useful when you need a quick summary versus a thorough analysis of the same input.
The second is Dynamic Workflows in Claude Code, currently in research preview. Claude can now spin up hundreds of parallel subagents to tackle large-scale tasks such as codebase migrations or multi-repository analysis. For engineering teams working with complex, distributed codebases, this compresses what used to take hours into something closer to minutes.
The third is an update to the Messages API: system-level instructions can now be inserted within the messages array, not just at the start. This means you can update context mid-task without breaking the prompt cache, a technical detail that matters for long-running agentic workflows where instructions evolve as the task progresses.
For sales teams tired of cold leads, slow customer responses, and manual processes, Dapta is the ultimate tool.
Dapta is the leading platform for creating AI sales agents specifically designed to increase inbound lead conversion. Respond to your leads in less than a minute with voice AI and WhatsApp that converts.
If you want your team to sell more while AI handles the complex stuff, you have to try it.
Claude Opus 4.8’s fast mode runs at 2.5x the speed of standard mode and is now priced at $10 per million input tokens and $50 per million output tokens. That represents a threefold reduction in fast-mode cost compared to previous generations. For teams deploying Opus at scale for tasks that do not require maximum depth, such as first-pass code review, document summarization, or customer support triage, this opens up a deployment tier that was not economically viable before.
The model is also more honest about uncertainty. It is less likely to make unsupported claims and more likely to surface its own limitations before presenting a result. Combined with tool-use improvements that require fewer steps to reach the same outcome, Opus 4.8 is designed for agents that operate with less human oversight, where the cost of silent failure is highest.
For businesses in Latin America and globally building AI-powered operations on top of Anthropic’s API, Claude Opus 4.8 makes a compelling case: better safety, lower cost at speed, and new infrastructure for large-scale parallel work, all at the same price point as its predecessor. The model is available immediately on all platforms and accessible via the API using the identifier claude-opus-4-8.
