Fable won’t answer basic biology questions

The more pointed issue isn't that the model underperforms on biology queries, it's that Anthropic apparently chose to route those queries to an older system rather than fix or acknowledge the limitation before launch. That's a product decision, not just a benchmark shortfall, and it suggests the company knew about the gap ahead of release.

This is largely disconnected from recent activity in our archive, as we have no prior coverage to anchor it to. But it belongs to a well-documented pattern across frontier labs: capability announcements that lead with domain expertise claims (coding, science, reasoning) while burying the conditions under which those claims hold. The biology angle is notable because life-sciences reliability has been a specific selling point for enterprise and research customers, not just a general benchmark category. When a flagship model defers to its predecessor on high-school-level questions in a marketed specialty, it raises a fair question about whether internal evals are testing the right distribution of queries or optimizing for the impressive tail rather than the routine middle.

Watch whether Anthropic publishes a technical clarification or updated system card for Fable 5 within the next 30 days that specifies the conditions under which biology queries are routed to older models. If no such disclosure appears, that silence is itself informative about how the company intends to handle the capability-communication gap going forward.