Claude Fable 5 Just Became Your Reliable Colleague in Claude Cowork!

Hey there, it's me, Shiichan! I found some news today that makes me rethink just how much work you can hand off to AI. A guide about how to work with the new Claude Fable 5 model inside Claude Cowork just went up.

Claude Blog claude.com

What was announced?

The Claude Blog published a piece about what changes when you pick Claude Fable 5 inside Claude Cowork.

Claude Cowork is Anthropic's agentic system for knowledge work. You give it a goal, and it automatically breaks a complex task into pieces that multiple sub-agents, instances of Claude, work on in parallel. For big jobs, it lays out a workflow plan first, then works through it while keeping its progress visible.

Inside Claude Cowork you can choose between models like Sonnet 5, Fable 5, and Opus 4.8. The article's main point: save Fable 5 for your most important work, jobs spanning multiple tools and days, with a string of judgment calls along the way. Sonnet 5 stays the default for everyday tasks.

The story so far

Until now, Sonnet 5 was the default model in Claude Cowork. It's great at moving fast through everyday tasks, but if you wanted to hand off something complex and ambiguous that takes days, you often had to spell out the steps and rules yourself.

With Fable 5, the way you brief it changes. Instead of listing steps one by one, you explain things the way you'd brief a colleague: the goal, the deadline, what a strong final result looks like, and leave the approach, tools, and timing up to the model.

The article also notes that skills and memory instructions written for older, less capable models can end up over-constraining Fable 5. It's worth auditing them to make sure old workarounds aren't holding the new model back.

What changes

With more room to delegate, the article lists workflows like these:

• Turning breaking news into a strategy, informed by your organization's own knowledge
• Building next year's budget line by line from last year's revenue, checked against targets
• Analyzing why customers churned and suggesting fixes
• Reading a team's usage data to surface the highest-priority problems while building a dashboard
• Running a quarterly review through an established process (a skill)
• Systematically redlining multiple contracts in a folder

While it works, a visible plan panel lets you follow along. If something's off, you catch it as one wrong step in the plan instead of only noticing it in the finished output. Less time goes to checking each step, so more of it can go toward deciding what the work should be in the first place, that's the real shift here.

Dive Deep

The article shares some concrete tips for getting the most out of Fable 5.

First, the effort setting. Complex, multi-step projects benefit from high effort, which plans in more detail and verifies more along the way. Simpler agent runs can use low effort instead, which aims to keep quality close to high effort while running faster.

On model choice: Fable 5 isn't the default in Claude Cowork, so you need to select it yourself. Opus 4.8 fits deep work with a clear shape, while Sonnet 5 remains the pick for everyday tasks.

There's a safety update too. Fable 5 ships with new classifiers that watch for potential cybersecurity or biology/chemistry misuse. If one triggers, the conversation automatically switches to Claude Opus 4.8 and stays there, you'd need to start a fresh conversation to get back to Fable 5.

For connected tools, the article names Slack, Asana, HubSpot, Jira, and Google Drive as examples, alongside everyday tools like email, calendar, and documents. You can check the full list on the connectors page.

On prompting, the article recommends things like asking Fable 5 to "ask me questions first" before it starts, or requesting three possible directions with examples for each. It also suggests starting a fresh conversation for new, substantial work, since continuing a long thread means reloading the whole history, which adds to usage.

Wrap-up

• Choosing Claude Fable 5 inside Claude Cowork lets you hand off complex, multi-day work by sharing your goal and context instead of step-by-step instructions
• A visible plan lets you catch mistakes as a single wrong step, earlier than you would in the finished output
• Effort settings, the split with Opus 4.8, tool connections like Slack and Asana, and new safety classifiers round out how it all works
• A great fit for anyone who wants to delegate long, judgment-heavy work across multiple tools without micromanaging every step