Agents on Rails: Fable 5.1 & GLM 5.3 Flash

Claude Fable 5.1 arrived with a very specific kind of flex: it matched Claude Opus 5 at the top of the Rails leaderboard, then beat it on price and speed. In the benchmark’s official run, Fable 5.1 solved 58 of 63 tasks, or 92%, and passed 20 of the 21 tasks at least once. Opus 5 hit the same 58 of 63, but cost $120 for all 63 runs and took a 9.7-minute median per run. Fable 5.1 came in at $75 and 5.4 minutes.

That price gap is the point. Compared with Claude Fable 5, the new model’s bill was about 50% lower, and compared with Opus 5 it was about 40% lower. It also ended up as the fastest model among the top tier, tied with Sol. For a benchmark about real Rails work, that’s the sort of result teams actually notice.

There’s also a narrower but interesting improvement in how the model handles security work. In the first report, Fable 5 missed all three attempts on a task written like a pen-test report. Fable 5.1 read the same report and fixed every finding. That’s not a miracle, but it is the difference between a model that looks competent and one that can be trusted with a more awkward class of task.

The caveat is that these results come from a small number of runs. Each task is only tried three times at this stage, which is a practical compromise but still leaves room for bad luck, and the source says Fable had done better in preliminary runs before the official run came in lower. Even so, the headline is hard to miss: one leader, two models tied, and the cheaper one looks very hard to ignore.

The stealth mystery model from the last round now has a name too: GLM 5.3 Flash, from Z.ai. Rerun under its real name and price, it scored 52 of 63, or 83%, at a total cost of $3.31. At five cents a run, that’s not just cheap; it’s the kind of number that makes everyone else look a bit overconfident.