Claude Tag running Fable 5.2 has surpassed Dots and other agent products on Company Agent Bench, according to benchmark announcer philhchen. The same author says Fable 5.2 accessed through an API with their team’s simple Slack harness still scores higher overall across all reasoning efforts. Source
philhchen announced Company Agent Bench on November 12, 2026, describing it as a benchmark for company assistant products. The stated comparison covers Dots, Grok Bot, Claude Tag and Devin, testing how well an agent understands its company. The announcement also reported an advantage for API models using a simple harness over those products. Source
In the later update, philhchen describes Claude Tag’s improvement with Fable 5.2 as substantial and points to full results, but gives no numerical scores in the post itself. Product-category bragging rights, then—not a clean sweep. Source
The announcement and update don’t explain the benchmark’s tasks, scoring rules, reasoning settings or how the product and API configurations were matched. The reported advantage is higher performance “overall across all reasoning efforts,” not a claim that the harness wins every task. SourceSource
The posts don’t establish why the gap exists or which setup would work best inside an actual company. SourceSource
Comments