Model chooser  /  Automate multi-step work with an agent  /  One-off

For a long, one-off task an agent should run end to end, use Claude Fable 5.1.

A one-off agent task, such as auditing a site, migrating data between tools or assembling a report from ten systems, fails on the long tail of steps. Fable 5.1 is Anthropic's model for long-horizon agentic work, and for one run its higher price barely registers.

How to use it ↓ Try another task

Why Claude Fable 5.1

The reasons it wins for this job.

  • Anthropic positions Fable 5.1 for "demanding reasoning and long-horizon agentic work". It is its most capable widely released model, launched September 2026.
  • It scores 55.8% on Terminal-Bench 4.0 against 52.3% for Opus 5, and shares the top Artificial Analysis Intelligence Index score (53).
  • Its thinking is always on, and it can keep working through long turns, which is what a task with dozens of steps and dead ends needs.
  • For a single run, cost per task is small in absolute terms. What you are buying is fewer restarts.

How to use it

4 steps to a first result.

  1. 1Write the task as an outcomeDescribe what finished looks like and how to verify it, not the steps. Over-prescriptive instructions make this model worse.
  2. 2Give it the tools and no moreConnect only the systems it needs, read-only where possible.
  3. 3Let it run, check progressExpect long turns. Ask for a short progress note at each milestone instead of approving every step.
  4. 4Verify the result yourselfCheck the output against the definition of done before anything is shared or sent.

Task brief for an agent

Start from this.

Edit the parts in capitals, then run it.

Goal: OUTCOME IN ONE SENTENCE.

Done means:
- CHECKABLE RESULT 1
- CHECKABLE RESULT 2

You have access to: TOOLS OR SYSTEMS. Treat them as read-only unless I say otherwise.

Do not: send emails, delete anything, change permissions, or spend money.

Work until done. At each milestone, write me two sentences on what you finished and what is next. If you are blocked for more than a few attempts, stop and tell me what you need.
Nothing is sent anywhere. It copies to your clipboard.

Alternatives that also work

If Claude Fable 5.1 is not an option.

GPT-6 AstraOfficial page →

OpenAI · closed · cost: high

Pick it when the task involves operating a computer or browser in ChatGPT or Codex. Astra leads Terminal-Bench 4.0 at 57.9%.

Claude Opus 5Official page →

Anthropic · closed · cost: high

Pick it when you want it faster and the task is well understood. Opus 5 is the default in Claude Code and Cowork.

Kimi K3Hugging Face →

Moonshot AI · open weights · cost: free weights; you pay for the hardware · 2.2M HF downloads/mo

Pick it when you want the strongest open-weight agent. Moonshot reports 91.2 on BrowseComp, but check its license limits.

Watch out for

Doing this again and again? For an agent that runs on a schedule, use Claude Opus 5. A production agent has to finish reliably, at a cost you can budget, without a human watching. Opus 5 sits two points behind the leaders on intelligence at half Fable's price, and Anthropic's Managed Agents can run it on a schedule with the sandbox hosted for you.

Sources

Checked . Models change monthly; we re-check this page when they do.

Picking the model is the easy part.

Wiring it into a workflow that runs every week, with evals, fallbacks and a cost you can predict, is the work. Fifteen minutes, no deck.

Book fifteen minutes →