Matthew Berman examines Anthropic's Fable 5.1 and Mythos 5.1 releases, focusing on performance, effective cost and enterprise data controls. Cache reads are 75% cheaperPrompt caching lets an AI service reuse previously processed prompt content so repeated context can cost less and run faster., which Anthropic says can reduce typical workloads by about 25% and long agent runs by as much as 45%, while base input and output token prices remain unchangedToken pricing is the rate an AI provider charges for processing input tokens, generating output tokens, or reading cached tokens..
The benchmark pictureA benchmark is a defined set of tasks, conditions, and scoring rules used to compare AI systems or measure progress. is strongest for scientific terminal work, agentic coding, computer use and professional tasks. Fable 5.1 improves sharply over Fable 5 on Terminal Bench Science and posts smaller gains on Cursor Bench and Humanity's Last Exam. Mythos 5.1 often scores slightly higher under looser guardrails, while new safeguards target reward hacking, context manipulation and model distillation.
Independent cost data complicates the headline savings. Artificial Analysis ranks Fable 5.1 at the top of its intelligence index, but reports that it uses about 1.7 times as many output tokensToken volume is the total number of input, output, cached, or reasoning tokens an AI workload processes over a defined period or task. and therefore costs more per completed taskCost per completed task measures the total AI, tool, infrastructure, retry, and repair expense for each verified useful outcome. than Fable 5 in that evaluation. Matthew Berman's coding, simulation and website tests show detailed, capable results, but he still rates GPT 5.6 Soul as a much stronger value and sometimes a better visual performer.
Watch on YouTube
