Research Papers Kimi K3 is an open Mixture-of-Experts model with 2.8T parameters (104B activated per token), native vision input and a 1-million-token context window, built on Kimi Delta Attention and post-trained with reinforcement learning across general, agentic and coding tasks.
FreeToken rethinks local inference around two realities of edge AI: agent workloads keep changing shape, and every machine's resource balance differs. It co-designs model layout, expert residency, CPU-GPU execution and memory management instead of committing to a fixed offloading strategy.
Prime Agent is an open harness for long-horizon coding agents built on the Recursive Language Model idea: a persistent IPython session does test-time computation, while histories, memories, skills and subagent specs carry over between runs.
Symbolic music models plan composition explicitly but never sound like a finished recording, while audio models produce full songs that keep the composition hidden. YuE2 combines both in a single AR-NAR Mixture-of-Transformers that plans symbolically, expands the plan into semantic music tokens, and renders the full song, using two new representation models (MERT2, SheetSage2) to learn from recordings that lack aligned scores. The authors report a SongBench Global Avg of 6.73 on WildSongBench (6.96 at best-of-8), and in expert listening the best-of-8 samples were preferred over Suno v4.5 and roughly tied with Suno v5; the readable score also makes the output editable by external language-model agents.
Xiaomi's robotics team presents a vision-language-action foundation model for mobile manipulation, pretrained on more than 100K hours of real-world UMI trajectories with auto-labeled scene-transition language, then post-trained to match embodiments and human imperative prompts. The paper claims clean scaling with more data and parameters, out-of-the-box performance in unseen environments and efficient fine-tuning for dexterous tasks.
VoiceMem gives duplex speech models a working memory that keeps facts and emotional context across a conversation, so a voice assistant can remember both what you said and how you said it, in real time.
Ouroboros is a self-developing coding-agent harness whose tools, prompts, context assembly and even core implementation improve through reviewed commits that become its later runtime, with two evolution modes: recursive free evolution and experience-driven core evolution. The authors also report a 161-day live deployment across seven human communication surfaces, arguing safety guardrails must stay authoritative while the agent rewrites its own code.
After Meta's closed Muse demo showed what a single-person agent with lasting memory could look like, this report reconstructs how Muse is built from public sources, then ships nanoMuse as its GPL-3.0 open counterpart: one agent across the devices you own, sharing a single conversation over a relay anyone can self-host. Every action passes a Sentinel gate before it executes, memory is stored as plain files the owner can read, and the underlying model is the user's choice.
Making a long film with a video-generation agent breaks down at the seams: passing prompts or a few recent frames between shots gives the model neither a shared spatial frame for revisiting a scene nor a time-indexed record for replaying an earlier event. ATAS compiles a screenplay into shot requests with explicit location, camera, and event-time references, and keeps two memories behind the generation loop: a scene memory of registered static geometry and an event memory of dynamic observations indexed by scene, event, and time. Each generated clip is reconstructed and registered back, admitted only after review and state checks. The authors report that this improves all measured cross-shot consistency metrics across four video generators, outperforms compared agents and 3D-guided methods on spatial revisits and event recapture, and earns the highest human ratings on all six evaluated criteria.
Frontis-MA1 ships with OpenMLE, an open full-stack system (task gyms, RL operator learning, evolutionary search) for studying recursive self-improvement. Its 35B meta-evolution agent, post-trained around Draft, Improve, Debug and Crossover program-evolution operators, claims MLE-Bench Lite results the authors say exceed GPT-5.5 plus Codex, with components transferring to held-out NatureBench Lite.
Raven argues that hand-crafted agent harnesses are too domain-coupled to scale with long-horizon, cross-domain workloads. It treats each model–harness pair as a composable unit of intelligence: a Host Agent decomposes goals and coordinates specialized sub-agents, a host archive and EverOS retain experience across tasks, and Skill Forge turns that experience into reusable procedures — with theory on expanding reliable task coverage under one resource budget.
ABot-World-0 is an action-conditioned video world model trained on multi-source data from AAA games, simulation engines and internet videos, controlled through raw keyboard input with reference-character memory for consistent third-person rollouts. The authors report streaming 720p at up to 16 FPS on one RTX 5090 with 1.2s action-to-first-frame latency, targeting real-time long-horizon closed-loop interaction.
Real agents get noisy tool outputs and vague user corrections, and most trained agents just barrel ahead. This paper from Renmin University builds a three-stage pipeline (RSD, ReSFT, ORPO) that first mines reflection behaviors from agent trajectories, distills them into an 8B base model, then sharpens proactive reflection with RL. Built on Qwen3-8B, the authors report gains across six agent suites and stronger robustness on their AgentNoiseBench, which injects realistic noise into tool feedback.
Edge0 is an open-source inference engine that serves a 35B mixture-of-experts model from a single 24GB GPU by loading expert weights from SSD on demand; a trained per-layer prerouter predicts the next layer's expert choices one token ahead so SSD reads overlap with compute, and a recovery LoRA pays back the quality lost to int4 quantization. The authors report 20 tok/s at about 3GiB of peak active memory, within a few points of the fp16 teacher on five public benchmarks.
This harness maintains an explicit task state outside the agent's execution context and updates it only with facts independently verified from the environment, using a Manage-Execute-Audit loop so mistaken self-assessments do not snowball across many steps.
Computer-use models can already read a screen and click, but games—with their visually rich worlds and diverse mechanics—remain hard, and pipelines built on large-scale demonstrations cover only a slice of dynamic game states. Player-1 post-trains such models with synthetic experience instead: it extends the computer-use action space with relative pointer motion, sustained inputs, and composite actions; fine-tunes on 6,044 curated gameplay trajectories paired with task instructions and visual reasoning; and then runs PPO in Player-1-Gym, sixteen synthetic 2D and 3D environments built around control demands shared across games, with programmatically verifiable rewards and randomized observations. The authors report that Player-1-9B and Player-1-35B-A3B beat their own SFT policies on MineDojo-300 by 27.8 and 17.7 percentage points, with Player-1-9B reaching 38.33% success versus 12.63% for the strongest baseline, plus gains on ViZDoom-10 and GameWorld—and the improvements transfer beyond the RL training suite.
This 66-page position-style paper formalizes harness self-evolution: an agent keeps its model weights fixed but continually revises its own software scaffolding, including prompts, memory, workflows, executable components, and plugins. Accepted revisions persist across runs, and the authors separate what can be updated from the proposal and admission mechanisms, analyzing reliability with PAC-learning-style bounds and an admission rule that requires fresh independent validation data. The paper argues this demands continual validation plus independent safeguards on runtime privileges and consequential actions.
Video understanding agents usually need slow chain-of-thought decoding, which is too late for tasks that demand timely choices from a video stream. VideoJev instead casts temporal grounding and multiple-choice video QA as structured decision tasks: it swaps the MLLM's autoregressive answer head for a shared scoring head that rates candidate time spans and text options against the video directly, trained with GRPO. A cached- and joint-inference scheme reuses visual computation across sequential queries on the same video. The authors report mean-mIoU gains of 6.7 and 10.8 points on 4B and 35B-A3B backbones with 2.8-3.1× grounding speedups, and over 100× speedups on video QA while matching Thinking-mode accuracy.
DiffusionART fixes two known weaknesses of online RL for diffusion models: reverse-process methods like Flow-GRPO waste compute on poor credit assignment, while forward-process methods like DiffusionNFT are fast but unstable and depend on carefully tuned EMA. It aggregates each prompt's rewards into one shared improving direction and adds a trust-region-masked restoration step, so it trains stably without a differentiable reward model or per-task tuning. The authors report that on HunyuanVideo-13B it matches the late-stage reward of DiffusionNFT and Flow-GRPO-Fast with roughly 12x fewer GPU hours, and the released framework covers T2I, T2V and I2V with six baselines for comparison.
OmniCompass evaluates video understanding along three complementary tasks: temporal grounding (when a target appears), spatial grounding (where it is), and question answering (what happens), in two tracks. OmniCompass-VL targets vision-language models with 820 videos and 2,460 instances, while an audio-rich subset forms OmniCompass-AV with 240 videos and 720 instances for models that must combine audio and visual evidence. Every instance is tested under paired text-only and text+image settings, and all data is manually annotated with multi-round verification. Evaluating 50 MLLMs, the authors report open-source models still trail proprietary ones, especially on audio-visual understanding, and that precise temporal and spatial grounding remains hard even for leading models.