ContextBench measures what an agent keeps, updates, and offloads when its input does not fit in its context window. Each task streams a sequence of operations as ordinary user turns, more than the context budget holds, and the reward is read from the The all config holds every instance. This release has 256 instances: 64 kvstore, 56 logtriage, 80 needleretention, 56 sudokusketchpad (32 levels, seeds 42-49), with a context budget of 32768 tokens and a reserve of 2048. Each operation is shown to the agent as === {header} ({remaining} op(s) remaining after this) === followed by its body. To evaluate an agent from Python, install the contextbench package, write an instance to disk, and pass…