Benchmarks / Agentic

PostTrainBench v1.1

Agents act as post-training engineers: given a base model and a compute budget, run a full fine-tuning pipeline, scored on downstream benchmark gains averaged over seeds.

11models

–Open Index weight

–last updated

PostTrainBenchpublisher

#	Model	Effort	Score	Cost / task	Harness
1	Fable 5	max	41.8%	–	–
2	GPT-5.6 Sol	max	36.2%	–	–
3	Opus 5	max	35.0%	–	–
4	Opus 4.8	max	33.8%	–	–
5	Kimi K3	max	32.0%	–	–
6	GLM 5.2	max	31.7%	–	–
7	Opus 4.7	max	28.6%	–	–
8	GPT-5.5	xhigh	27.2%	–	–
9	Grok 4.5	high	23.4%	–	–
10	Gemini 3.1 Pro	high	22.0%	–	–
11	GPT-5.4	high	19.0%	–	–