AI Coding Live News

On Oct 27 Liquid AI released LFM3, a new generation of foundation models built on a tri-hybrid architecture of gated short convolutions, sliding-window attention and global attention, pre-trained on 22T tokens. The first six models all branch from LFM3-1.7B-Base at around 2B parameters, with weights on Hugging Face under the LFM Open License v1.0. Liquid says the text model LFM3-1.7B needs about 1 GB at 4-bit, supports 256K context and beats models several times its size on several agent benchmarks.

Zoom

Researchers introduced MemPilot (arXiv:2610.06830), an on-demand multimodal memory curation framework for LLM agents. Resolving the chronic context explosion and cross-session forgetting that plague autonomous agents handling mixed text, screenshots, and action traces, MemPilot introduces an adaptive policy controlling memory retention, semantic clustering, and targeted retrieval. Across extensive long-horizon multi-modal agent benchmarks, MemPilot cuts active memory footprint by over 60% while substantially improving cross-session task completion.

Zoom

On-policy distillation (OPD) is critical for reasoning and agent alignment, yet conventional methods realize only a single continuation per decision, severely restricting supervision under finite compute budgets. Researchers have introduced BranchOPD (Branching On-Policy Distillation), an algorithm that dynamically identifies high-uncertainty and teacher-student disagreement decision nodes to branch and resample continuations from shared prefixes. BranchOPD slashes generation costs by 19%–31% on agent benchmarks while boosting math reasoning accuracy by 18.55 percentage points.

Zoom

On Oct 13 Liquid AI released LFM2.5-24B-A2B, the largest and final LFM2.5 model: 24B total parameters, ~2.45B active per token, 131K context, and always-on reasoning. By Liquid's own numbers it leads or matches Qwen3.6-35B-A3B on 5 of 8 tool-use/agent benchmarks (AutomationBench 27.69 vs 8.72) while trailing on SWE-Bench Verified, GPQA and other knowledge/coding tests. Open weights are on Hugging Face with GGUF, MLX, FP8 and ONNX builds; API pricing is $0.10 / $1.00 per million tokens and it is also on OpenRouter.

Zoom

Mainstream coding agent benchmarks like SWE-bench frame evaluation around issue-derived tasks: feeding an agent a bug description and expecting an end-to-end patch. But does this reflect genuine developer practices? JetBrains Research analyzes 4,782 production sessions of professional software engineers inside JetBrains IDEs (arXiv:2610.09633). For long interactive sessions (>=3 turns, comprising 33% of production traffic), user requests span a diverse mix of intent types—project codebase Q&A, strategic planning, code review, refactoring, and test execution—with users frequently transitioning between types mid-session. To align benchmarks with real developer task flows, JetBrains introduces SWE-TaskFlow, an approach that transforms issue-based benchmarks via prompt-splitting and verifiable repository QA into calibrated multi-stage workflows guided by a TaskFlow Alignment Score (TFAS). A pilot study across 700 SWE-Bench Pro tasks demonstrates that sequential multi-step interaction protocols nearly double agent inference expenditure without stable improvements in resolve rates, proving that the interaction protocol itself is a critical, previously unmeasured dimension of agent evaluation.

Zoom

Whether LLM-based autonomous agents truly learn from interactive experience or merely regurgitate memorized knowledge remains a fundamental question in AI research. Most conventional agent benchmarks evaluate tasks (such as writing standard Python or navigating operating systems) heavily represented in pretraining corpora, creating severe data contamination that obscures whether agents are synthesizing new concepts or retrieving memorized patterns. Researchers introduce Learn2Play-Bench (arXiv:2610.08215, code: github.com/liushiliushi/Learn2Play-Bench, site: liushiliushi.github.io/learn2play-bench-website), a novel open-source benchmark constructed around text-based environments with entirely unfamiliar and counterintuitive mechanics. Agents must discover game physics and causal dependencies purely through trial-and-error interactions. The benchmark reveals two critical insights: retaining uncompressed, verbatim interaction logs decisively outperforms LLM-generated rule summaries, and the agent harness scaffolding (managing tool loops, memory, and reflection) exerts a larger impact on final performance than the choice of underlying foundation model.

Zoom

Reinforcement learning (RL) pipelines for code agents typically rely on binary test suite outcomes for reward assignment. Under standard Group Relative Policy Optimization (GRPO), all test-passing rollouts within a sampled group receive identical advantage values, completely ignoring code quality, extraneous diff modifications, and algorithmic elegance. This induces severe trajectory bloat and training instability. Researchers introduce GAGAR (Groupwise Agentic Grading and Advantage Redistribution, arXiv:2609.32577). In a shared workspace, an agentic evaluator jointly inspects all passing candidates in a group, ranking them by architectural clarity and minimal intervention. A sum-preserving redistribution reallocates advantage weights toward superior implementations. Validated at industrial scale on MiMo-V2.6-Flash (310B) and MiMo-V2.6-Pro (1.02T parameters), GAGAR curtails trajectory length explosion while boosting code agent benchmark performance and optimization stability.

Zoom

Coding agents frequently reproduce existing code, error logs, and multi-turn refactoring traces, making retrieval-based speculative decoding (SD) an ideal acceleration paradigm. However, conventional retrieval drafters fail in agent workflows: candidate code on disk diverges from the agent's emission formats (e.g., diff blocks, escape characters), and static draft lengths ignore inter-agent turn drift. KAIST researchers introduce AgSpec, a speculative decoding framework tailored for coding agents. AgSpec constructs format-aligned corpora across session trajectories, workspaces, and global repositories, coupling offline draft caps with online feedback-adaptive draft length control. Across repository-level multi-agent benchmarks, AgSpec outperforms five retrieval drafters and EAGLE-3, boosting generation throughput up to 4.76x.

Zoom

Overcoming the limits of traditional multi-agent benchmarks that only test short-horizon (<20 steps) or competitive settings, UPenn researchers released AgentWorld. Spanning 50+ interaction rounds in an MMORPG sandbox across 100 human-annotated tasks with 3-20 asymmetric black-box agents, it introduces Causal Collaboration Effectiveness (CCE). Even frontier models achieve only 52.0% task success, exposing severe communication breakdown and shared plan decay.

Zoom