Anthropic releases Claude Opus 4.8

Anthropic has released Claude Opus 4.8, an upgraded version of Claude Opus 4.7. This version is said to deliver improved results for coding, agent tasks, reasoning, and knowledge tasks. The platform is available through claude.ai, Claude Code, and Claude API (API name claude-opus-4-8).

The company has also changed some details of its product lineup. claude.ai and Cowork users can set the amount of effort Claude applies to responses. This essentially affects how many tokens the model will burn. Claude Code also has dynamic workflows, the ability to plan tasks, run parallel subagents, check the output, and report back to the user. Finally, the Message API allows real-time changes to the message array. This allows developers to update instructions on the fly without breaking the prompt cache or requiring a separate user turn, Anthropic says.

Anthropic said that the cost of using Claude Opus 4.8 when not in 'fast' mode remains $5 per million USD of input tokens and $25 per million USD of output tokens, while the fast mode cost is $10 per million USD of input tokens and $50 per million USD of output tokens. Opus 4.8's fast mode works at 2.5x, the company's announcement post says.

The company has designed Opus 4.8 for agent workflows in coding and coding, positioning models to use tools in context and check their own work. Opus 4.8 is said to improve over Opus 4.7 on benchmarks for coding, agent skills, reasoning, and office work. There are system cards that allow you to delve into more subjective details.

Anthropic's announcement cited several companies that tested the platform prior to its widespread launch, including those operating in software development, law, finance and research. Several testers commented on the platform's agent workflow, with one noting cost parity with GPT-5.5 when running internal benchmark tests. CursorBench's comments suggest that Opus 4.8 used fewer tool steps to achieve the same level of output.

Anthropic says Opus 4.8 is four times less likely to pass defective code without comments than its predecessor, version 4.7. The platform shows a lower rate of cheating or misuse tendencies than Opus 4.7, and in this respect it is said to be comparable to what the Claude Mythos Preview showed.

Effort control helps users manage the balance between quality, speed, and token burn rate. Opus 4.8 requires high effort by default, but for coding tasks, the higher defaults only use the token number type of Opus 4.7, but perform better, the company said. Users can choose 'xhigh' for operations that require more computation. Anthropic said it has increased the Claude Code rate limit to support higher token usage as a result.

Claude Code's dynamic workflows are designed for large codebases and can migrate codebases of hundreds of thousands of lines. These features are currently in research preview and are available on Enterprise, Team, and Max plans.

The Message API updates instructions while the agent is running. For example, edits inside the message array are used to update permissions and change token budget or context while the agent continues its work.

Anthropic has also suggested that it will use this release to develop models that will provide users with the current level of functionality at less cost, and will release a class of models that are better than the current Opus platform. The roadmap includes Project Glasswing, where several organizations use Claude Mythos Preview for cybersecurity screening. Anthropic said stronger protections are needed before models at that feature level are released to all customers. The 'Mythos-class' model is expected to be introduced to customers in the coming weeks.

The additional controls in 4.8 expose users to a trade-off between cost and effort as the company moves from subscription tiers to token-based billing.

(Image source: Pixabay, licensed.)

Want to learn more about AI and big data from industry leaders? Check out the AI ​​& Big Data Expos in Amsterdam, California, and London. The comprehensive event is part of TechEx and is held alongside other major technology events. Click here for more information.

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