Research Papers Rankings
The overall board: every AI paper from the last 90 days, across language models, agents, vision, robotics and the rest. Ranked by three independent signals — GitHub star velocity on the official implementation, citation count, and Hugging Face attention — so a paper that only trended for a day does not hold a place. Rebuilt every day.
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.
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.
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.
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.
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.
AutoDesign treats multimodal conference-poster generation as a long-horizon agentic process: a meta-harness optimizer reads rollout feedback and rewrites the harness the code agent runs on. On the authors' 100-paper PosterBench it scores above Claude Design, the learned harness lifts seven different code-agent configurations, and a fully autonomous loop makes 253 tool calls in 40 minutes for under $3.
Alaya-EVOKE externalizes persistent scene geometry into a camera-indexed world state bank so the denoiser context stays bounded, and trains with a sparse-attention teacher that supervises long horizons at linear cost, exposing content drift and supporting prompt changes mid-sequence. The teacher distills into a three-step student that the authors report producing each 1.5-second chunk in 2.11s on a single H200 GPU.
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.
Most video benchmarks score a single prompt-to-clip pipeline, but streaming models must keep generating while new instructions arrive mid-stream. StreamEval scripts 300 interactive test cases (15-300s each) across five tasks: avatar dialogue, avatar performance, world generation, instruction editing, and reference editing, where multimodal requests arrive incrementally and each case demands both state changes and content preservation. Scoring 37 models on 10 metrics across four dimensions, the authors report that strong visual quality does not translate into following instructions: the best request-execution scores are only 36.90 (avatar performance Following), 25.93 (world generation Following), and 15.00 (reference editing Edit Success) out of 100, flagging reliable real-time request handling as the field's main open problem. Leaderboard at streameval.ai.
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.
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.
Ad-hoc data mixing and model-specific implementations give video world model research inconsistent supervision and hard-to-reproduce comparisons. SolarWM open-sources the whole pipeline: a reconfigurable data engine unifying 1.43 million canonical clips from 10 datasets, four 5B-33B instantiations on Wan2.2, LTX-2.5 and MiniMax-H3 behind common conditioning, training and inference APIs, and a three-stage recipe combining bidirectional adaptation, teacher-forced autoregressive initialization and distribution-matching distillation. The authors report that causal models trained on only 5-second sequences become world models capable of real-time interaction over rollouts lasting minutes to hours.
MOSS-Image is an open generation system aimed at the gap where open-weight image models still struggle with dense text and intricate layouts. It pairs a 5B single-stream diffusion transformer pretrained from scratch, which the authors say enables bilingual long-form text rendering and precise layout control, with an intention-aware agentic prompt enhancement module that turns loose user requests into detailed content and layout specifications. The report introduces two new benchmarks, LongLongTextBench and RealLayoutArena, and reports that MOSS-Image is state of the art among open-weight models on long-form English and Chinese text rendering with 96.3% average text coverage, and ranks second in end-to-end layout-rich generation. The authors say they will release the model, training recipe, and evaluation framework.
Robot trajectory data is expensive and sparsely covers the physical world, so continued pre-training under a fixed budget tends to just fit actions rather than build transferable knowledge. VLAct pre-trains a VLM backbone on broad multi-embodiment robot data with multi-head continuous action co-supervision and a partially unified cross-embodiment action layout, preserving the VLM prior and re-attaching task-specific action heads at fine-tuning time. The authors report 82.6% and 92.5% success rates on LIBERO-Plus and RoboTwin 2.0, and on the unseen RoboCasa-GR1 humanoid their model beats the full-data GR00T-N1.6 baseline using only 20% of downstream trajectories — trained entirely on open-source data with a 16-GPU setup.
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.
Fine-grained perception usually means stitching together separate specialist models for detection, segmentation, counting, and OCR. WeVisOne-VL (from the WeChat Vision team) unifies object-level and pixel-level tasks in one VLM: a lightweight pixel decoder plugged in via a special <OBJ> token lets the same model output boxes, points, masks, and even alpha mattes at whatever granularity the user asks for. Trained with three-stage curriculum SFT plus mixed-task RL, the authors report leading results across 30+ benchmarks while still refusing queries where the requested object is absent, a failure mode they explicitly target for real-world deployment.
OpenGCUA rebuilds the computer-use agent stack in the open, motivated by the observation that frontier systems like GPT-6 Astra's computer use are closed. Its core idea is to let the model choose between visual GUI actions and direct code execution, so intermediate results can steer later steps; it also proposes harness-to-model decision transfer, moving selected observation and task-state decisions out of fixed harness rules and into the model itself. The authors report that on OSWorld-v2 their full configuration improves task scores across all eight evaluated models (including GPT-6 Astra and Claude Opus 5), raising mean partial score by +19.6 points and cutting reported steps per task by 53.0% on average versus batched GUI-only execution. Alongside the implementation and prompts they release evaluation trajectories from all eight models, plus early versions of a real-device harness and a cross-device SDK covering computers, mobiles, and browsers; on four real macOS tasks their harness shows selected Qwen3.8-Max runs use 19.4 times as many reported reasoning tokens as GPT-6 Astra, which they highlight as a gap in real-device execution efficiency.
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.
Scaling laws are well established for language models in the physical world, but do they hold for embodied agents acting in virtual worlds with different graphics, physics, controls, and objectives? This paper runs a controlled study of continued pre-training across three axes, with 120 configurations spanning four model sizes (0.8B to 27B), six data budgets (50 to 1,600 hours), and five data mixtures from Minecraft-only to diverse multi-world corpora. The findings: more data helps within the training worlds, data diversity transfers to related worlds but barely to distant ones, and larger models use data more efficiently yet generalize worse to distant worlds under single-world training. Training also splits into an early interface phase with broad cross-world gains and a later world phase where gains narrow to familiar worlds. The authors fit an empirical scaling law relating validation loss to volume, diversity, and size, and note that persistent gaps in distant worlds mean generalist embodied agents will need training stages beyond continued pre-training.
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.
The overall board: every AI paper from the last 90 days, across language models, agents, vision, robotics and the rest. Ranked by three independent signals — GitHub star velocity on the official implementation, citation count, and Hugging Face attention — so a paper that only trended for a day does not hold a place. Rebuilt every day.
A dash means the signal was unavailable, not zero.