AI Coding Live News

When software engineers interact with LLM coding agents, natural language requirements are frequently ambiguous, underspecified, or missing critical boundary rules. Existing coding agents face a difficult trade-off: silently hallucinating assumptions to produce syntactically valid but functionally flawed code, or incessantly asking trivial clarifying questions that disrupt developer focus and cause cognitive fatigue. Researchers from Peking University and Wuhan University introduce CONTRA (arXiv:2610.01769, repo: github.com/fangz-cs/Contra), a training-free framework for selective clarification in LLM code generation. CONTRA employs a three-tier pipeline: broad question discovery, semantic- and execution-based question qualification, and history-aware adaptive stopping. By executing candidate code variants in isolated environments conditioned on alternative answers, it verifies whether a question triggers stable, behavior-changing differences before prompting the user. On ClarifyCodeBench, CONTRA achieves macro-average question F1 of 41.20% and TKQR of 38.32%, outperforming leading baselines by 13.88 and 12.33 percentage points respectively. On GPT-5.5, it lifts recall from 15.12% to 36.72% while maintaining high precision. CONTRA is released as an open-source extension for Claude Code.

As autonomous coding agents demonstrate proficiency in software debugging and code generation, researchers are advancing them into discrete 3D spatial engineering and physical assembly. Researchers from Stanford University and Inria introduce BrickBench and the BrickAgent interactive environment (arXiv:2610.12452, website: brickben.ch), formulating the first comprehensive benchmark for agentic, text-conditioned LEGO set design. Given natural language prompts, coding agents must select parts from discrete modular libraries and generate executable Python code defining 3D coordinates, orientations, and interlocking poses. Generated assemblies are formally scored across structural validity, text-design alignment, and aesthetic complexity under rigorous physics-based simulation. Evaluating frontier models reveals that while leading coding agents can satisfy verifiable physical and semantic requirements, they consistently fall short of human design elegance and structural efficiency, establishing a pivotal milestone for spatial coding agents.

Reinforcement learning (RL) has emerged as the defining post-training paradigm driving frontier foundation models toward autonomous self-improvement, yet scaling RL across multi-step agent environments and million-token sequences introduces profound stability and infrastructure hurdles. Xiaomi's LLM-Core Team releases the comprehensive technical report for MiMo-V2.6 (arXiv:2610.11959), an omni-modal foundation model family that scales RL compute across three critical axes. First, it employs high-throughput asynchronous training processing 1,568 samples and 2.7 to 3.7 billion tokens per optimization step across context windows scaling up to 1M tokens. Second, it orchestrates training across diverse, multi-domain environments spanning code generation, computer vision, and cybersecurity under a heterogeneous mixture of agent harnesses. Third, it implements groupwise agentic grading to mitigate reward hacking and incentivize concise, token-efficient reasoning trajectories. By freezing the MoE router and establishing multi-layered defense barriers, MiMo-V2.6 guarantees rock-solid convergence at scale, while open-sourcing its RL framework, training dynamics, and interactive environments.

Large language models frequently encounter a probability mass dilemma in multi-step reasoning: while the correct sequence might possess a higher individual probability than any single wrong candidate, the vast ocean of incorrect paths collectively dominates the posterior distribution. Sequence-level power sampling sharpens distributions toward the mode by raising probabilities to exponents greater than one, but requires generating and scoring dozens of rollout candidates at runtime. Researchers from USC and collaborators introduce On-Policy Power Distillation (OPPD, arXiv:2610.06804), an algorithm that teaches models to output power-sharpened solutions in a single forward pass without inference-time search. By running an on-policy Sequential Monte Carlo (SMC) teacher-student distillation loop, OPPD boosts single-generation accuracy by 23.0 points on MATH500 and 27.3 points on GSM8K without external reference solutions. Crucially, OPPD outperforms verified-reward GRPO on MATH500 (+3.8), GSM8K (+4.0), and AIME (+5.4) under identical compute budgets, while cross-generalizing to raise HumanEval code generation by 5.3 points. The repository is available at github.com/ArminAzizi98/OPPD.

OpenRouter announced Anthropic’s Claude Fable 5.1 is live on its routing platform as a direct upgrade for Fable 5 workloads. The biggest gains are called out in agentic coding, long-running workflows, visual code generation, finance, and analysis, and developers can call it via OpenRouter immediately.