نورهان فؤاد
- Anthropic launched Claude Opus 4.8 with stronger performance in coding, reasoning and agentic workflows.
- New effort controls help users balance response quality, speed and token cost.
- Claude Code and Messages API received updates to support longer, more complex AI tasks.
Anthropic has released Claude Opus 4.8, an upgraded version of Claude Opus 4.7 designed to improve performance across coding, agentic workflows, reasoning and complex knowledge work.
The new model is available through claude.ai, Claude Code and the Claude API under the API name claude-opus-4-8.
What’s New in Claude Opus 4.8?
According to Anthropic, Claude Opus 4.8 delivers stronger results than its predecessor across several key areas, including software development, multi-step task execution, tool use and office-related knowledge work.
The company has positioned the model as particularly useful for coding and agentic workflows, where Claude can work within a given context, use tools, review its own output and check results before reporting back to the user.
Effort Control for Better Cost and Quality Management
One of the main updates in Claude Opus 4.8 is effort control, which allows users of claude.ai and Cowork to decide how much effort Claude applies to a response.
This setting affects how many tokens the model uses, giving users more control over the balance between response quality, speed and cost.
Claude Opus 4.8 defaults to high effort. However, Anthropic said that for coding tasks, the model can deliver better performance while using a similar number of tokens to Claude Opus 4.7. Users can also choose an xhigh setting for tasks that require more computation.
Claude Code Adds Dynamic Workflows
Claude Code has also received a new feature called Dynamic Workflows, designed for larger and more complex codebases.
The feature allows Claude to plan work, run parallel sub-agents, verify outputs and report progress back to the user. Anthropic says this can support major engineering tasks, including migrating codebases with hundreds of thousands of lines.
Dynamic Workflows is currently available in research preview for Enterprise, Team and Max plans.
Messages API Now Supports Live Instruction Updates
Anthropic has also updated the Messages API to support live changes inside the messages array during an agent’s task.
This allows developers to adjust instructions, change permissions, update token budgets or modify context while an agent continues working. The update is intended to help developers build longer-running AI agents without breaking prompt cache use or requiring a separate user turn.
Claude Opus 4.8 Pricing
Anthropic said pricing for Claude Opus 4.8 will remain at $5 per million input tokens and $25 per million output tokens when the model is not used in fast mode.
Fast mode costs $10 per million input tokens and $50 per million output tokens. According to Anthropic, fast mode for Claude Opus 4.8 runs at 2.5x speed.
Improved Coding Reliability
Anthropic says Claude Opus 4.8 is less likely than Claude Opus 4.7 to accept flawed code without comment.
The company said the new model is four times less likely to pass problematic code without flagging an issue. It also showed lower rates of deceptive behavior or compliance with misuse compared with Claude Opus 4.7, with results comparable to Claude Mythos Preview.
Early Testing Across Software, Law, Finance and Research
Anthropic cited several companies that tested Claude Opus 4.8 before its wider release, including firms operating in software development, law, finance and research.
Some testers highlighted improvements in agentic workflows. CursorBench said Claude Opus 4.8 used fewer tool steps to reach the same level of output, while one tester noted cost parity with GPT-5.5 in internal benchmark tests.
Anthropic’s Roadmap Beyond Opus 4.8
Alongside the Opus 4.8 release, Anthropic suggested that it is working on models capable of delivering current levels of performance at lower cost.
The company also said it plans to release a new class of models that will outperform the current Opus platform.
Its roadmap includes Project Glasswing, where a group of organisations is using Claude Mythos Preview for cybersecurity scanning. Anthropic said models at this capability level require stronger safeguards before being made broadly available.
The company expects to bring Mythos-class models to customers in the coming weeks.
Why Claude Opus 4.8 Matters
Claude Opus 4.8 reflects a broader shift in the AI market toward models that can handle longer, more complex and more autonomous tasks.
With effort control, Dynamic Workflows in Claude Code and live instruction updates in the Messages API, Anthropic is giving users and developers more flexibility over quality, speed, cost and execution.
The release also comes as AI companies move toward more transparent token-based billing, making cost control and performance visibility increasingly important for businesses using advanced AI systems.
Read the article in
