Anthropic released Claude Fable 5.1 a few hours ago. It says the model is better at coding and at long-running problem solving, and that at low or medium effort it matches or beats Claude Fable 5 at a much lower cost on the API. We graded it the same day at medium effort, as the direct successor to Claude Fable 5 on the bench.
What the results show
- It is the best agent on the bench: Claude Fable 5.1 solved more than the rest of the field, and it is the first agent to earn an overall B on our leaderboard.
- It is faster everywhere except planning. Against Claude Fable 5 the median real-world issue took 1m 46s instead of 2m 37s, and a vibe-coding fix 3m 48s instead of 5m 28s. On planning it took longer, 6m 51s instead of 5m 32s, but that time was not wasted, because the plans it wrote scored better.
- It spends less of the plan usage when coding and more of it on thinking. On the same tasks a coding task took 1.8% of the Claude Max 20x five-hour window on average, against 2.3% for Claude Fable 5. A planning task took 3.8% against 2.5%, which is where you want the extra spend to go.
Claude Fable 5.1 is a clear upgrade over Claude Fable 5. It knows where to invest more effort and where to invest less: it codes quicker and better, and it spends more time planning. Fable is not really a model to run constantly but something for the hardest problem, where it has taken another step in the right direction. This solidifies the Claude subscription plans as the best option once Fable is included, which starts at the $100 tier.
Grades are comparable within the same suite version. See the full Claude Fable 5.1 scorecard or follow the family over time on Trends.