Topic
Techniques applied after pretraining—fine-tuning, RL, distillation—to refine an LLM's behavior, alignment, and agentic capabilities.
Epoch AI's InnovationEval credited GPT-5.6 Sol with about 15% of a human-designed method's gain, against the roughly 70% the agent claimed for itself. Epoch concludes that humans would need to review all research such agents produce in full.
Reality
Z.ai says the base model did not change between GLM-5.2 and GLM-5.3, so the coding jump and the doubled exploitation score come out of the same post-training run. Security teams inherit the second half.
Perspective Coverage
4 publishersReality
Zhipu says every gain in GLM 5.3 came from post-training on the same ~744B MoE it shipped as GLM 5.2. The largest claimed leap, in vulnerability discovery, is also the least quantified.
Reality
ServiceNow's CoreAI team generates new agent training tasks, each with its own verifier, aimed at gaps a stronger teacher model can already solve. Adopting it means running a seedable copy of the environment and writing verifiers that accept any valid solution.
Reality
Z.ai says every gain in GLM-5.3 came from post-training on an unchanged base. If that holds, refresh cadence for self-hosted weights is set by RL runs, not pretraining runs.
Perspective Coverage
5 publishersReality
An arXiv paper names the failure behavioral state decay and runs a side-car memory agent next to an unmodified action agent, reporting gains of 8.3 and 6.8 points of pass@1 on two long-horizon benchmarks.
Reality
Harvey and Baseten report rubric pass rates rising from 23.3% to 62.4% across seven models once the data room sits in a Python REPL and sub-agents do the reading, while Claude Code with Opus-5 manages 24.6%.
Reality
An empirical study of publicly released PostTrainBench trajectories finds the training strategy is fixed at step one and the whole remaining budget goes on local tweaks, with scaffolding and human hints improving execution and leaving that pattern intact.
Reality
Microsoft reports Qwen3.5-9B going from 41.8% to 56.4% on SWE-bench Verified using 6K examples. The load-bearing change is who owns the agent loop, and what the service boundary costs.
Reality