Reasoning

RLHF can make agents persuasive assistants without making them dependable autonomous decision-makers. Builders should separate human-pleasing interaction from calibrated automation and keep stakes bounded.

Training-data curation can improve model quality and inference efficiency without simply adding compute. The practical work is decontamination, deduplication, balancing, task matching, and selective synthesis.

Base-model data is shifting from broad web imitation toward code, reasoning, and agent-task priors. The unresolved choice is how early to introduce synthetic and instruction-shaped data.

Safety tuning against model self-consciousness may also shift unrelated value and mind-attribution responses. Builders using models for surveys or social reasoning should treat alignment as a confound.

β-OPSD exposes self-distillation’s fixed regularization as a tunable parameter, then approximates policy optimization through logit mixing. It targets more stable reasoning training without direct RL.

Inkling Small is pitched as a lower-compute model for coding, tool use, and visual reasoning, with adjustable thinking effort and zero-data-retention routing through Vercel AI Gateway.

Grok Voice Think Fast 2.0 brings speech-to-speech reasoning and earlier tool calls to Vercel’s realtime API, with server-minted tokens keeping gateway keys off clients.

OpenAI positions GPT-5.6 as delivering more useful output per dollar across inference and agent workflows. The supplied material has no metrics for judging routing or migration decisions.

Skill-SP turns agent skills into units for verifiable self-play: generate tasks, solve them, then update the skill library from execution feedback. The abstract provides no per-benchmark effect sizes.

CausalForge pairs a Lean-verified causal-inference library with an autonomous research pipeline and a semantic statement audit. Formal proof checks derivation, not whether the theorem matches the intended claim.

This security eval tests whether agents can discover and exploit logic flaws across live chained services, using hidden zero-days and deterministic grading instead of source-code pattern matching.

VLM-IE3D adds implicit and reconstructed geometry tokens to an RGB-video VLM, offering an open approach for agents that must reason about spatial scenes without dedicated 3D input.

MIRROR trains a VLM across text, diagram, and combined views by letting its strongest view supervise weaker ones, targeting the modality inconsistency that single-view evals hide.

X³-OPD transfers a text model’s reasoning into an audio-language model while grounding training in the student’s own acoustic interpretations, including events, prosody, and dialogue.

Experimental on-device agents can play games and adapt interfaces without cloud calls, but real-time use must fit memory, frame-time, and battery budgets. Accessibility is promising, not production-ready.

This survey maps how LLMs inspect and regulate their reasoning, giving agent builders a framework for choosing self-checks without assuming introspection is reliable.

A low-dimensional theory links training data and initialization to whether transformers reason through context or learned weights, but only on a generalized synthetic task class.

AdvancedMathBench separates proof writing from verification and finds frontier models especially weak at rejecting invalid proofs, a warning against trusting agent self-review on rigorous reasoning.

Agora routes reasoning steps through an auction among expert models and tools, adding a single control for cost versus quality and outperforming matched baselines on five benchmarks.

Vercel’s limited preview exposes GPT 5.6 as Sol, Terra, and Luna, giving coding-agent teams flagship, balanced, and lower-cost routing targets behind one gateway.

A free, code-oriented Chinese curriculum spans model tuning, deployment, agents, alignment, security, and multimodal systems. It is useful as a broad learning map, but remains a work in progress.

Direct-OPD reuses a small model's RL run to improve a bigger one: the pre/post-RL log-ratio becomes a dense reward for the stronger student, lifting Qwen3-1.7B from 48.3% to 62.4% on AIME 2024 in 4 hours on 8 A100s.

DemoPSD gates self-distillation per token by teacher–student disagreement, cutting the answer-leakage shortcuts that hurt generalization; beats GRPO and SDPO on science QA in and out of domain.

DramaSR-532K benchmarks speaker attribution over 532K dialogue lines and 900+ TV-drama characters; a reasoning LLM with multimodal tool use beats acoustic baselines, especially on short utterances.

ReContext is a training-free harness that replays query-relevant evidence from long inputs before answering, taking the best average rank across 8 long-context benchmarks up to 128K on Qwen3-4B/8B and Llama3-8B.

Google's AMIE matched 21 primary-care physicians on longitudinal disease management in a blinded Nature study, scoring higher on plan preciseness and guideline alignment. Research-stage, not deployed.

OpenAI introduced GPT-5.6 with claims of improved token efficiency and cost-performance, but supplied no measurements or access details for model selection.

Grok 4.5 extends Cursor’s model pool to long-running tool work beyond coding, but its CursorBench result is excluded because an earlier Cursor code snapshot entered training.

Grok 4.5 is available through Vercel AI Gateway with text and image input plus low, medium, and high reasoning settings for tuning speed against depth.