reasoning benchmark
PostTrainBench Lite
PostTrainBench Lite measures whether an agent can design and execute a full post-training strategy (data, prompts, RL recipe, and eval loop) for a pretrained base model under a constrained time budget, scored as normalized mean reward over the improvement window.
Updated Aug 7, 2026
Published models3
Registry coverage3
MetricScore
EvidenceB
PostTrainBench Lite leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | OP | 51.5% | 100.0% | 3 | C | |
| 2 | OP | 50.3% | 50.0% | 3 | C | |
| 3 | OP | 29.6% | 0.0% | 3 | C |
Score distribution
Top published rows on the benchmark's original scale.
PostTrainBench Lite highlights
The top published results on this benchmark's own scale.
What is PostTrainBench Lite?
Definition and scoring fields from the benchmark registry.
PostTrainBench Lite measures whether an agent can design and execute a full post-training strategy (data, prompts, RL recipe, and eval loop) for a pretrained base model under a constrained time budget, scored as normalized mean reward over the improvement window.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- PostTrainBench Lite
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- posttrainbench-lite|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about PostTrainBench Lite.
Which model scores highest on PostTrainBench Lite?
GPT-5.6 Terra is currently ranked first with 51.5%.
What does PostTrainBench Lite measure?
PostTrainBench Lite measures whether an agent can design and execute a full post-training strategy (data, prompts, RL recipe, and eval loop) for a pretrained base model under a constrained time budget, scored as normalized mean reward over the improvement window.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
3 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.