Good evening, fam... A thousand AI agents built an operating-system prototype. The cost curve is the part I would watch. Market Pulse | 10 Nov 2026 Anthropic's November 9 release of Fable 5.2 and Mythos 5.2 gives us two very different pieces of evidence. The chart shows Anthropic's FrontierCode 2.0 results. Fable 5.2 at High effort scores 45.3%, close to Opus 5.5's best 45.5%, at roughly half the cost per task. This benchmark assesses whether code changes would be accepted by open-source maintainers. Look further along the green line: Xhigh reaches 49.1%. Max costs more and scores lower in this evaluation. Paying for more reasoning does not automatically buy a better result. The demonstration in my reply shows Anthropic's operating-system prototype. The company says Fable coordinated over 1,000 subagents across eight days to build it. That illustrates the scale of work it is attempting. The video and benchmark cover different workloads, so the chart cannot tell us what that OS project cost. The practical question is whether cheaper model execution also means cheaper finished work. Retries, integration and human review can absorb the saving. Over the next 72 hours, I will look for independent tests that report accepted results, total spend and reviewer time on the same tasks. Lower cost with comparable quality would strengthen the case. Savings that disappear during checking would weaken it. That is the number I want from an agent: the full bill for work I can actually use. Chart: Anthropic. Benchmark results are company-reported. https://t.co/dSLUQCBRR5