The Arrival of Claude Fable 5.1
Anthropic has just released Claude Fable 5.1, alongside a sibling model, Mythos 5.1. According to the company's announcement, the release "sets a new standard for coding, knowledge work, and long-running problem-solving tasks."
The data backs up this claim, particularly in complex domains. On the newly released Terminal-Bench-Science 0.1 benchmark, Fable 5.1 scored 52.6%. To put that in perspective, its predecessor, Fable 5, scored just 24.7%. Anthropic's Opus 5 achieved 29.0%, and OpenAI’s GPT-5.6 Sol lagged at 22.4%. While other benchmarks showed only slight improvements, this massive jump on a rigorous science benchmark points to a fundamentally stronger engine for deep problem-solving.
The Cost of "Thinking" Time
Developer Simon Willison put Fable 5.1 to a practical test, utilizing his standard benchmark: asking the model to generate an SVG image of a pelican riding a bicycle. His breakdown reveals the most critical operational feature of the new model. Fable 5.1 operates on five distinct reasoning levels: low, medium, high, xhigh, and max. Notably, there is no option to turn reasoning off entirely.
The performance gap between these settings is staggering, both in completion time and financial cost:
- Low effort: Produced a result in 23.8 seconds, generating 1,998 tokens for roughly 10 cents.
- High effort: Took 29.6 seconds and 2,612 tokens, costing 13 cents. The model generated a brief internal summary of its layout plans.
- Max effort: Took 13 minutes and 54 seconds, generating 65,927 output tokens at a cost of $3.30.
At the "max" setting, the model's reasoning trace was vast. It actively debated design constraints, noting that a bicycle helmet might "compete with the crest for visual space" and adjusting control points on the bicycle's front fork to fix the "rake." The result was a highly detailed, accurate SVG. Willison then piped this complex output back into the model on the "high" effort setting with a simple "animate this" prompt, generating a working, animated SVG for an additional $1.37.
What Fable 5.1 Means for SMB Automations
For small and mid-sized businesses automating their operations, Fable 5.1 introduces a new architectural lever: the explicit trade-off between compute time, cost, and reasoning depth.
Historically, automation platforms treated AI steps as near-instantaneous transactions. You send a prompt; you get an answer. Fable 5.1 shifts advanced automation into the realm of asynchronous, long-running batch processes.
This requires a strategic split in how businesses build AI into their internal workflows. For high-volume, low-complexity administrative tasks—like extracting line items from daily vendor invoices, categorizing inbound customer support tickets, or parsing lead data—automations should be locked to "low" effort. You get reliable, formatted data in seconds for pennies per transaction.
However, operations that require deep reconciliation or complex logic can now be handed off to the "max" setting. If a business needs to reconcile a month’s worth of messy, mismatched inventory records across three different software platforms, a standard fast-response automation will likely fail. But configuring a system to hand that data to Fable 5.1 on "max" effort changes the economics of the back office. Paying $3.30 and waiting 14 minutes for the model to deeply reason through edge cases, self-correct its logic, and output a clean reconciliation report is vastly cheaper than the hours of human labor it would otherwise require.
Effective AI automation is no longer just about stringing together instant prompts. It is about matching the precise reasoning effort to the operational value of the task. Fable 5.1 proves that when you give the system the time and token budget to think, it can execute deep, multi-step knowledge work that was completely out of reach just a short time ago.