Claude Opus 4.8 Released: What's New in Anthropic's Latest AI Model

Anthropic has officially released Claude Opus 4.8, a significant upgrade to Claude Opus 4.7, delivering measurable improvements across coding, agentic workflows, reasoning, and knowledge work. The model is accessible via claude.ai, Claude Code, and the Claude API under the identifier claude-opus-4-8.

🔧 Key Platform Updates

Anthropic has also refined its product lineup with several notable additions:

• Effort Control: Users of claude.ai and Cowork can now set how much effort Claude applies to a response — directly influencing the number of tokens the model consumes.
• Dynamic Workflows in Claude Code: A new feature that plans work, runs parallel sub-agents, verifies outputs, and reports results back to the user — ideal for large-scale codebases.
• Messages API Live Updates: The Messages API now accepts real-time changes to the messages array, allowing developers to update instructions mid-task without disrupting prompt cache usage or requiring a separate user turn.

💰 Pricing Overview

🔹 Standard Mode: $5 per million input tokens / $25 per million output tokens
🔸 Fast Mode: $10 per million input tokens / $50 per million output tokens
⚡ Fast Mode operates at 2.5× speed compared to standard mode.

🤖 Built for Agentic & Coding Workflows

Anthropic has positioned Opus 4.8 as a model purpose-built for coding and agentic tasks, with the ability to use tools within a context window and self-verify its own work. Benchmark results show improvements over Opus 4.7 across coding, agent skills, reasoning, and office productivity tasks. A full System Card is available for deeper technical review.

In its official announcement, Anthropic highlighted early adopters across software development, law, finance, and research. Feedback from testers emphasized the model's agentic capabilities, with one noting cost parity with GPT-5.5 on internal benchmarks. CursorBench reported that Opus 4.8 required fewer tool steps to achieve equivalent output quality.

🛡️ Safety & Reliability Improvements

Opus 4.8 is four times less likely than Opus 4.7 to pass flawed code without flagging issues. It also demonstrates lower rates of deception and is less susceptible to misuse — performing comparably to Claude Mythos Preview in this regard.

⚙️ Effort Control & Token Management

The new Effort Control feature helps users balance quality, speed, and token consumption. Opus 4.8 defaults to high effort, yet on coding tasks, this default consumes a similar number of tokens as Opus 4.7 — while delivering superior results. For compute-intensive tasks, users can select the xhigh setting. Anthropic has correspondingly increased Claude Code rate limits to accommodate elevated token usage.

Dynamic Workflows within Claude Code are engineered for large codebases — capable of migrating repositories with hundreds of thousands of lines of code. This feature is currently in research preview and available exclusively on Enterprise, Team, and Max plans.

🌍 Anthropic's Roadmap: Project Glasswing & Beyond

Anthropic signaled its longer-term ambitions alongside this release — including the development of models that deliver current capability levels at reduced cost, as well as a next-generation model class superior to the current Opus platform.

Its roadmap includes Project Glasswing, a program in which select organisations are already using Claude Mythos Preview for cybersecurity scanning. Anthropic noted that models at this capability level require stronger safeguards before broader release, and expects to bring "Mythos-class" models to customers in the coming weeks.

The expanded controls in Opus 4.8 are also part of a broader strategy as Anthropic transitions from subscription-tier billing to token-based billing, giving users greater visibility into cost and effort trade-offs.

📷 Image source: Pixabay, under licence.

📣 Want to learn more about AI and big data?

Check out the AI & Big Data Expo — taking place in Amsterdam, California, and London. The event is part of TechEx and co-located with other leading enterprise technology events. Click here for more information.

AI News is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.