GLM-5.2 Ponytail

Ponytail is documented by GreenPT as an open-source token-compression skill aimed at coding agents, not as a standalone model release. In the provider's compression overview, it sits between two companion skills: Honey, which combines tighter code with terse prose and agent handoffs, and Caveman, which is reserved for terse prose. Ponytail is positioned as the minimal-code, YAGNI-oriented option in that trio, helping agents emit less code while keeping the output functionally intact, and the page frames the whole family as a way to lower per-request cost, latency, and energy use in line with GreenPT's sustainability goals.

For practical fit, the documentation advertises the skill as delivering roughly 54% fewer code tokens with up to 94% retention, while running about 20% cheaper and 27% faster on GreenPT's chat completion surface, where compression variants can be selected by model id and have their own system prompt that merges with any prompt a caller supplies. The compressed output is intended to remain technically accurate, and the skill is offered alongside Honey and Caveman so teams can match compression intensity to the task, from aggressively terse prose to leaner code without sacrificing correctness on agent workflows.