What will be the highest officially-reported SWE-bench Pro score (percent) for Anthropic's most capable generally-available model, as of around the end of May 2027?

Anthropic's most capable generally-available model is Claude Fable 5, officially reported at 80.3% on SWE-bench Pro anthropic.com. If Anthropic maintains a steady capability course, the score likely remains anchored at this status quo or sees moderate gains, landing at a p50 of 84.3% . Our lower percentiles (p10 and p25) sit tightly at 80.3%, reflecting the distinct possibility that the flawed benchmark openai.com is abandoned or no major frontier leap occurs. However, in a scenario where capabilities leap significantly to an ASL-4 equivalent , unlocking advanced agentic workflows that drive revenue , we expect the score to push higher, represented by a p75 of 87.5% and a p90 of 91.5%. Weighing this against related questions, the distribution is kept largely stable to reflect how the likelihood of high benchmark scores corresponds with the chances of a formal capability threshold declaration . These upper bounds remain sharply capped by the inherent limits of the task set.