Anthropic Claude Opus 4.8 Ships Side by Side with Powerful Workflows and Cheap Fast Mode, with Workflows Set to 1,000 Subagents

Anthropic recently released Claude Opus 4.8. Also, there are two Claude Code updates posted. A dynamic workflow uses multiple subagents in parallel. Quick mode now supports Opus 4.8 at a lower price. Both are preliminary research studies.

What Dynamic Workflow Really Is

A dynamic workflow is a JavaScript script that schedules subagents at scale. Claude writes a script for the job you describe. The runtime then runs in the background. Your time remains responsive while the agents are working.

The program enters the code, not Claude's context window. Intermediate results live in script variables instead. So Claude's context has only the last answer. That is the main difference from subagents and abilities.

The feature requires Claude Code v2.1.154 or later. Works with CLI, Desktop, and VS Code extensions. Available in Max, Team, and Enterprise plans. It is automatically enabled for Max and Team. In Enterprise it is disabled until an administrator enables it. It also works on Claude API, Amazon Bedrock, Vertex AI, and Microsoft Foundry.

You can start a workflow in two ways. Enter a name workflow anywhere in your knowledge. Or turn on the setting called ultracode. Ultracode includes xhigh an attempt to think about the orchestration of automated workflows. Claude's code also includes /deep-research as a built-in application.

How the Application Works

When the workflow starts, Claude dynamically edits from your input. It divides the task into sub-tasks. Its followers work in all subagents that work in parallel. Agents approach the problem from independent angles. Other agents then tried to dispute the findings. The run repeats until the answers converge. The results are checked before they are rolled.

The runtime uses strict limits. It allows up to 16 simultaneous agents. It covers each run for a total of 1,000 agents. The workflow script itself cannot touch the file system or the shell. Only agents read, write, and execute commands.

Progress is saved as the run continues. An interrupted task resumes during the same session. Terminated agents return cached results on restart. Communication happens without a conversation, so the system is always on track.

Example of Rewriting

Anthropic highlighted one great example. Jarred Sumner implemented a flexible workflow to port Bun. Port moved Bun from Zig to Rust. It passed 99.8% of the existing test suite. It produced about 750,000 lines of Rust. It took eleven days from the first dedication to assemble.

A single workflow maps the appropriate Rust health to every field in the structure. The next person wrote each .rs file as port like behavior. Hundreds of agents worked in parallel, with two reviewers per file. The repair loop then calls the structure and checks until it is clean. The result has not yet produced.

What is Quick Mode Actually?

Fast mode is not a separate model. Claude Opus high speed configuration. Delivers 2.5x faster token withdrawal speeds. The model's intelligence and skills remain the same. You change it too /fast in Claude Code. A small one ↯ icon marks the active time.

Fast trade mode costs speed. With Opus 4.7 and 4.6, it is priced at $30/$150 per MTok. According to the official release of Anthropic, the fast mode of Opus 4.8 is three times cheaper. Requires usage credits enabled on your account. It comes from those credits, not your program's installed usage. It's best for quick iteration and live debugging. Normal mode is still equivalent to long tasks that are independent or sensitive to costs.

Power and what to watch

Power:

• The workflow handles codebase-wide bug hunting, mass migrations, and back-end audits.
• Non-combination reviews aim for results that a single pass cannot achieve.
• Quick mode maintains Opus-level quality while cutting interactive workflow lag.
• Restarting means that the interrupted workflow does not start over.

What you can watch:

• Both features use logically more tokens than a normal session.
• A run can spawn 1,000 agents, so the costs add up quickly.
• Fast mode costs from the first token to a higher value.
• Both ship as research previews, so pricing and availability are subject to change.

Marktechpost Visual Explainer

# 1. Trigger a workflow with the keyword
Run a workflow to audit every API endpoint under src/routes/ for missing auth checks

# 2. Let Claude decide when to use one
/effort ultracode

# 3. Use the built-in research workflow
/deep-research What changed in the Node.js permission model between v20 and v22?

Key Takeaway

• Anthropic recently released Claude Opus 4.8, which features two updates to Claude Code: a more powerful workflow and a cheaper express mode.
• The powerful workflow allows Claude to write orchestration scripts that use tens to hundreds of parallel subagents, deployed up to 16 at a time and 1,000 in total per run.
• The program lives in script variables, not Claude's context, so only the final response returns to your session.
• Fast mode uses Opus at 2.5x output speed with the same quality; it's now three times cheaper in Opus 4.8 and requires usage credits enabled.
• Both go as a research preview and consume more meaningful use, so start cutting and validating the output.

Check it out The project again Technical details. Also, feel free to follow us Twitter and don't forget to join our 150k+ ML SubReddit and Subscribe to Our newspaper. Wait! are you on telegram? now you can join us on telegram too.

Need to work with us on developing your GitHub Repo OR Hug Face Page OR Product Release OR Webinar etc.? contact us