Researchers are introducing step-level rewards, entropy-based objectives, and gradient-based quantization to make LLM reasoning and retrieval more efficient and reliable. These methods target token cost, length, and quality without always scaling model size.
A benchmark on 14 cross-system tasks and three MCP servers shows that TrueForge with Opus 4.8 solves 11/14 tasks like Claude Managed Agents, but uses 63% fewer tokens and costs 30% less per run. With GLM-5.2 the cost drops to $3. Native tracing, sand...
PRO-Step introduces step-level supervision for Retrieval-Augmented Generation. Instead of evaluating only the final answer, a generative process reward model checks logical validity and evidential grounding at every step of multi-hop reasoning. The m...
A new post-training quantization method, REAL-Q, uses block-wise gradient descent to reduce end-to-end KL divergence by up to about 49% compared with second-order methods on LLaMA-3.1 and Qwen3 at W4A16. The result matters for resource-constrained de...
TPGC introduces a dual prompt initialization for multi-task graph pre-training: a task prior and a structural prior extracted from an auxiliary graph. Across six node and graph classification benchmarks, the method outperforms baselines in few-shot s...
ERR+ introduces a two-phase RLVR framework that rewards token-level entropy drops and relative length efficiency. Correct traces show more frequent and larger entropy drops during thinking. Tests on five datasets report improved accuracy and concisen...
