AI Coding Live News

Researchers from NUS and CUHKSZ have introduced SolarRenderer (arXiv:2610.28630), a structure-conditioned generative reference rendering framework. Converting game engine grey-boxes, pose skeletons, semantic maps, and depth passes into photorealistic video, SolarRenderer adapts a frozen MiniMax-H3 33B backbone via lightweight control and 4-step generation LoRAs. Using structure-guided distribution matching, it renders stylized video in just 4 denoising steps (~12.3x faster than 50-step diffusion) and introduces a block-causal variant for real-time 768p streaming synthesis in generative game engines.

Zoom

The Shanghai Artificial Intelligence Laboratory published the technical report for Sage-A1.5 on arXiv:2610.28578, introducing an open-weights 35B Mixture-of-Experts (MoE) foundation model purpose-built for agentic tasks. Building on the Agents-A1 lineage, Sage-A1.5 focuses on scaling the agent execution horizon rather than model parameters through adaptive capability composition. Tailored for complex repository-level software engineering and multi-step scientific hypothesis exploration, Sage-A1.5 consistently outperforms dense 70B+ baselines on long-horizon reasoning benchmarks while maintaining lightweight deployment footprints.

Zoom

On Oct 20, NVIDIA's GEAR lab (Jim Fan's team) open-sourced Flora, a model suite that turns human and robot manipulation video into timestamped action captions, process rewards, outcome rewards and rubric scores. It includes Flora-Captioner 27B and Flora-Rewarder 27B (both on Qwen3.5-27B) and Flora-TempSeg 9B (a LoRA on Qwen3.5-9B), with weights on Hugging Face under an NVIDIA non-commercial license. NVIDIA says it is the in-house data engine used to train GR00T, labeling 200,000 hours of video a day on 512 GB300s, matching GPT-6 Astra on annotation quality at roughly 125x lower cost.

Zoom

Researchers from UC San Diego and Georgia Tech introduced ACT3D (arXiv:2610.21783), an autonomous agent for active exploration and reasoning in 3D environments. Rather than relying on heavy multi-view image suites, video streams, or 3D point cloud encoders, ACT3D preserves native monocular VLM encoding (processing a single first-person frame at a time like humans) while treating 3D space as an interactive environment. Optimized via outcome-driven RL without trajectory supervision, ACT3D achieves 53.2% average accuracy across six 3D benchmarks (+22.0% over base Qwen), outperforming proprietary Gemini 3.6 Flash while using 50%-70% fewer input frames.

Zoom

Modern Mixture-of-Experts architectures increasingly combine fine-grained expert segmentation with shared experts. However, ETH Zurich and HKU researchers identified an 'Exploration Dilemma' (arXiv:2610.22353, GitHub: BingcongLi/MirrorMoE): shared experts homogenize token representations, suppressing routing exploration just as fine-grained segmentation combinatorially enlarges the routing space. MirrorMoE resolves this via token-adaptive suppression of unnecessary experts, cutting pretraining perplexity across all model scales.

Zoom

Three new studies tackle verification, evaluation, and RL alignment for rule-governed generative world models: OmniVerifier-C1 (arXiv:2610.22355, 32 upvotes) uses Decoupled RL to train a decomposed graded visual verifier that reaches 0.684 on ViVerBench and boosts FLUX.1-dev GenEval from 0.68 to 0.79 while enabling Maze and Sudoku world modeling; Meituan's WorldCast benchmark (arXiv:2610.20731, 30 upvotes) evaluates 12 models across 320 multi-step world state prediction cases, finding GPT-Image-2.5 System at 72.8 vs. 42.8 for Nano Banana Pro and 20.1 for the top open model; and NUS's Routing Policy Optimization (RPO, arXiv:2610.22282, 23 upvotes) turns attention routing into an internal exploration action to train video/image diffusion RL 3.1x–4.2x faster than Flow-GRPO and DanceGRPO.

Zoom

Addressing rollout homogenization and zero-variance advantage collapse in group-relative reinforcement learning (GRPO), SERO (arXiv:2610.19397) restores learner-visited states to execute multi-step oracle continuations with policy-reachability weighting—boosting 7B ALFWorld OOD success by +11.20 points to 87.50%—while UNSW's EDEN (arXiv:2610.16105) appends a single anchored Gaussian-perturbed embedding position to a fraction of rollouts, lifting Qwen2.5-Math-7B AIME 2025 Pass@256 by +13.33 points (to 66.67%) without increasing rollout budgets.

Zoom

On 2026-10-10 Logesh Kumar Umapathi published GUI-Decisions on Hugging Face: computer-use actions/coordinates are read from fixed next-token slots instead of autoregressive JSON. On an RTX PRO 6000 with vLLM, Gemma-4-31B slot reads median ~146–158 ms/step vs ~472–546 ms native JSON (~3.5×) and ~20–25× vs thinking decode. ScreenSpot click-in-element ~0.87. Open weights for 31B/2B, GitHub, and a Spaces demo.

Zoom

As autonomous software engineering agents and command-line assistants evolve from single-turn tool invocation toward complex, multi-stage delivery, the scarcity of high-quality, long-horizon terminal interaction data has emerged as a fundamental bottleneck. Researchers from USTC, Alibaba Group (AMAP), and SUSTech introduce TerminalHorizon (arXiv:2610.13166), a fully automated data synthesis engine for training long-horizon terminal agents. Addressing standard pitfalls where synthetic environments oversimplify core challenges and eliminate the need for exploration, TerminalHorizon extracts workflows from real-world tasks, pinpoints omitted steps through independent execution loops, dynamically reconstructs environments based on observed evidence, and organically scales task complexity across multiple stages under unified objectives. Synthesizing 1.5K environments and 2,852 trajectories averaging 52.8 turns, the authors train TerminalHorizon-27B (fine-tuned on Qwen3.5-27B). The model achieves 55.8% on Terminal-Bench 2.0 (+14.2 points over base Qwen3.5-27B), sweeping 12 benchmarks across coding agents, production environments, and reasoning with project datasets open-sourced.

Zoom

Interactive virtual environments allow embodied AI agents to learn through exploration and practice, but creating worlds that are simultaneously faithful (rigid state dynamics and programmatic rules) and realistic (photorealistic visual distributions) has remained a critical bottleneck: physics engines lack visual fidelity, whereas generative video models suffer from compounding visual hallucinations during extended agent interactions. MirroS Lab and researchers introduce AgentGarten (arXiv:2610.12374, repo: github.com/MirroS-Lab/AgentGarten, site: mirros-lab.github.io/agent-garten), an open-source framework that decouples physics simulation backends from a shared neural renderer. Simulators maintain ground-truth state and execute code-defined rules without ingesting rendered frames, strictly preventing error accumulation. Visual observations are synthesized in real-time (>30 FPS) from exported geometric conditions via a neural renderer trained with Adversarial Forcing, making history prefilling differentiable through exact replay. Agents learn through visual perception, distilling execution trajectories into playbooks that succeeding generations inherit. In benchmarks such as hide-and-seek, agents autonomously discover obstacle-moving and ramp-climbing tactics in just 4 to 10 rounds (compared to tens of millions of rounds in classic RL), with code and environments fully open-sourced.

Zoom

Computer-Use Agents (CUAs) designed to operate graphical desktop environments through virtual keyboard and mouse interactions are traditionally evaluated using outcome-based functional verifiers that assess only the final state after hundreds of operational turns. This black-box approach obscures whether failures stem from syntax typographical errors, GUI bounding-box misalignments, or intermediate state drift, preventing actionable diagnostic insights. NVIDIA Research introduces OSWorld-Pro (arXiv:2609.24890), the first comprehensive process-based evaluation suite for computer-use agents. Spanning 300+ complex desktop workflows decomposed into 2,800+ sequentially dependent subgoals grounded in 67,000+ expert human annotations, OSWorld-Pro deploys calibrated LLM judges to track incremental subgoal completion. Evaluations reveal that OSWorld-Pro presents a formidable challenge to state-of-the-art models: frontier models like Claude Opus 5 achieve only 75.7% (compared to 83.4% on standard OSWorld), systematically isolating critical procedural failure patterns such as exploratory irrelevant actions and click grounding drifts.

Zoom