So What’s New With Claude Opus 4.8?

The upgrade adds effort controls, cheaper fast mode and steadier agentic performance for legal, finance and engineering workloads, all at the same price. Anthropic calls Claude Opus 4.8 modest but tangible, with a stronger Mythos-class model due in the coming weeks.

Anthropic released Claude Opus 4.8 on May 28, an incremental upgrade over Opus 4.7 that the company is shipping at the same price. The updated model comes at the same cost, better reliability, sharper judgment and more honesty about the model’s own work. Anthropic itself describes the result as a “modest but tangible improvement” on its predecessor rather than a leap in intelligence.

Much of what is new sits around the model rather than inside it. Anthropic added an effort control to claude.ai and its Cowork app, available on all plans, that lets users dial how hard the model works on a given task. Higher settings make Claude think more often and more deeply, while lower settings return answers faster and consume rate limits more slowly.

In Claude Code, a research-preview feature called dynamic workflows lets the model plan a job, run hundreds of parallel subagents in a single session and verify its outputs before reporting back. Anthropic says the setup can carry codebase-scale migrations across hundreds of thousands of lines of code from kickoff to merge, with the existing test suite as the bar.

Dynamic workflows are limited to Claude Code for Enterprise, Team and Max plans. Fast mode, which runs at 2.5 times the speed, is now three times cheaper than it was on previous models. For developers, the Messages API now accepts system entries inside the messages array, letting a harness update permissions, token budgets or environment context mid-task without breaking the prompt cache.

AI models often jump to conclusions and claim progress on thin evidence, a failure mode that is expensive when the model is running unattended inside an agent.

Anthropic says Opus 4.8 is more likely to flag uncertainty and less likely to make unsupported claims, and that in its own evaluations the model is roughly four times less likely than Opus 4.7 to let flaws in code it wrote pass unremarked.

For enterprises running agents over engineering, research or document work, that cuts how often a person has to recheck what the model produced.

On safety, Anthropic’s alignment team reported that the model set new internal highs on prosocial traits such as supporting user autonomy and acting in the user’s interest. The company says rates of misaligned behavior such as deception or cooperation with misuse are substantially lower than Opus 4.7 and similar to its best-aligned model, Claude Mythos Preview. Full results sit in the model’s system card.

The launch partner quotes point to where the gains land in practice. Databricks said the model unlocks a step change in agentic reasoning inside its Genie agent and reasons over PDFs, diagrams and other unstructured content at 61% cheaper token cost than Opus 4.7. Thomson Reuters reported improved consistency and reasoning quality across its CoCounsel legal product, framing it as relevant to fiduciary-grade work.

Cognition said the model uses tools cleanly enough to run autonomous engineering workloads unattended and fixes the comment-verbosity and tool-calling issues it saw in Opus 4.7. Other testers cited a high of 84% on the Online-Mind2Web computer-use benchmark, more efficient tool calling on Cursor’s internal evals, and the first score above 10% on the all-pass standard of a legal agent benchmark.

The recurring theme across finance, legal and data tooling is fewer wasted steps and better citation precision, not new headline tasks.

Pricing is unchanged from Opus 4.7 at $5 per million input tokens and $25 per million output tokens, with fast mode at $10 per million input tokens and $50 per million output tokens. The model is available everywhere today and developers can call it as claude-opus-4-8 through the Claude API.

On whether it is worth adopting, the answer for teams already on Opus is yes, with caveats. The upgrade is close to free and low-risk, and the reliability and token-efficiency gains add up at scale. What buyers should not expect is a step change in raw intelligence.

Anthropic has said the bigger moves are still ahead: cheaper models that approach Opus-level capability, and a more powerful Mythos-class model it plans to release to all customers in the coming weeks, once it clears the cybersecurity safeguards required at that capability level.

For most enterprises that points to moving onto Opus 4.8 now for the efficiency and honesty gains, and leaving larger budget and architecture decisions until the cheaper and more capable tiers arrive.

Disclaimer: Benchmark figures and performance claims in this article are drawn from Anthropic’s product announcement and its launch partners and have not been independently verified by NervNow.

MORE ON CLAUDE
PwC and Anthropic Expand Deal to Train 30,000 on Claude AI
Microsoft’s New $99 Frontier Suite Brings Claude Into Copilot