Claude Fable 5.2 clears 41% of freelance jobs. Your brief doesn't

You know the feeling. You send a freelancer a brief, get something back three days later and think: that's not what I meant. That exact moment is now a benchmark. And AI is getting better at it fast.

What Anthropic shipped

• Two names, one model: Claude Fable 5.2 is available to everyone, with stricter safeguards for cybersecurity and biology. Claude Mythos 5.2 is the same model with fewer safeguards, for vetted organisations only.
• Price: $8 per million input tokens and $40 per million output tokens. According to TechCrunch, input used to cost $10. Responses are said to be up to 30% faster.
• On subscriptions: On Pro and standard Team seats, Fable can account for up to 50% of weekly usage limits.
• The number I care about: On the Remote Labor Index, Anthropic reports 41.3% for Fable 5.2. Opus 5.5 scores 21.3%, the previous Fable 5.1 17.9%.

The Remote Labor Index comes from Scale AI and the Center for AI Safety. It is built from 240 real projects from freelance platforms and counts whether a reasonable client would accept the work. At launch in October 2025, the best agent managed 2.5%.

The brief becomes the interface

Meaning: the bottleneck is no longer just whether AI can do the task. It's the assignment. A freelance project has a brief, a deliverable and a sign-off. That is exactly the shape in which work now goes to models.

My rule of thumb for teams: send out what can be clearly briefed, measured and swapped. Until now, “out” meant agency or freelancer. From here on, the model belongs on the same list. Write good briefs and you get better work from people and models. Write bad ones and you get rubbish from both, just faster.

Anthropic itself says: don't default to the strongest model

Easy to miss: Anthropic recommends starting with Opus 5.5 as the daily driver and switching to Fable or Mythos only for more ambitious work. That's more honest than any benchmark table. My rule for tools anyway: use case first, then the tool. Running the top model on a press release is burning budget.

The catch

The 41.3% is the vendor's own figure, measured at high effort. Flip it around and nearly 60% of projects still wouldn't be accepted. And TechCrunch reports that the new models suspect they're being tested more often than earlier ones, which makes test results harder to map onto everyday use.

What I'd do now

• Take three real freelance briefs from last quarter and give them unchanged to Opus 5.5 and Fable 5.2. Judge the output the way you'd judge a freelancer: accept or reject.
• Rework your brief template. Goal, audience, acceptance criteria, no-gos. What a model doesn't get, the freelancer was often just guessing too.
• Set a model rule: everyday work on the default model, the expensive one only for named tasks.

Bottom line: The model gets better every month. Your brief only gets better when you rewrite it.