Benchmark RadarAI BENCHMARK PROFILE
PostTrain Bench
PostTrainBench evaluates a model's ability to autonomously post-train base models. Given pretrain-only base models, the agent must complete the full pipeline of data synthesis, training, evaluation, and iteration within a time budget, scored across downstream benchmarks such as AIME2025, BFCL, GPQA Main, GSM8K, and HumanEval.
- Released
- Unknown
- Readiness
- Paper only
- Primary field
- General AI
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.