Issue #6
Sent 2026-06-15
ABot-Earth 0.5: Generative 3D Earth Model
Reported FID of 16.1 versus 69.5 for EarthCrafter (the prior best baseline) on 2D renderings, with KID 0.006 vs 0.061. Major caveat stated by the authors themselves: baseline FID/KID were computed against different ground-truth sets and different camera poses, so the numbers are 'for reference only' rather than a controlled comparison.
arXiv:2606.09967 · 486▲ · cs.CV
- Vision
- Robotics
ABot-Earth 0.5 is a generative 3D system that synthesizes city-scale outdoor environments directly in the 3D Gaussian Splatting (3DGS) representation, conditioned on ordinary satellite imagery. It trains a compression-generation model on real-world 3DGS reconstructions (built by their own ABot-3DGS pipeline) and generates tiled scenes with a sliding-window inference scheme and native multi-level-of-detail output for streaming.
Full breakdown →Kwai Keye-VL-2.0 Technical Report
The report claims state-of-the-art results among similar-scale models on fine-grained temporal localization (TimeLens: ActivityNet, QVHighlights, Charades) and long-video benchmarks (Video-MME-v2, LongVideoBench), and reports top scores on tau^2-Bench (82.6) and VitaBench (33.1) for agentic tool use, with second place on BFCL-V4 (65.7). All figures are self-reported and several comparison columns in the tables are blank, so cross-model comparison is incomplete.
arXiv:2606.10651 · 192▲ · cs.CV
- Multimodal
- Efficiency
- Agents
A technical report on Kwai Keye-VL-2.0-30B-A3B, an open-source Mixture-of-Experts multimodal model (30B total, 3B active) that adapts DeepSeek Sparse Attention (DSA) to a GQA-based backbone so it can process 256K-token contexts, aimed at hour-long video understanding. It also describes a post-training recipe combining Cross-Modal Multi-Teacher On-Policy Distillation (MOPD) with Context-RL and Video-RL to add agent, code, and search skills without wiping out base reasoning.
Full breakdown →Embodied-R1.5: Evolving Physical Intelligence via Embodied Foundation Models
Reported state-of-the-art on 16 of 24 embodied VLM benchmarks with a 70.4% average across 21 accuracy benchmarks, versus 53.4% for Gemini-Robotics-ER-1.5. As a VLA it reports 92.4% on SimplerEnv Google Robot Visual Matching (over 20 points above pi-0.5). Caveat: results are self-reported through the authors' own evaluation framework, and the real-robot experiments are qualitative demonstrations rather than quantified success rates.
arXiv:2606.11324 · 170▲ · cs.RO, cs.AI, cs.LG
- Robotics
- Multimodal
- RL
Embodied-R1.5 is an 8B-parameter vision-language model fine-tuned to unify three robotics-relevant capabilities (spatial reasoning, task planning/correction, and pointing/trajectory generation) in one model, trained via supervised fine-tuning followed by reinforcement learning with verifiable rewards on 15B tokens of mostly auto-generated data. It also ships a Planner-Grounder-Corrector loop where a single model instance plays all three roles for long-horizon tasks, and can be extended into a robot action policy (VLA) by attaching a flow-matching action head.
Full breakdown →EvoArena: Tracking Memory Evolution for Robust LLM Agents in Dynamic Environments
Current agents average only 39.6% accuracy across the three EvoArena domains. Adding EvoMem gives a modest average step-accuracy gain of 1.5% and a 3.7% chain-accuracy gain on EvoArena, plus 6.1% on GAIA and 4.8% on LoCoMo; note the step gains are small and per-model, so the headline value is the benchmark exposing the failure mode more than the memory fix eliminating it.
arXiv:2606.13681 · 142▲ · cs.CL
- Agents
- LLMs
EvoArena is a benchmark suite that turns static agent benchmarks (terminal workflows, software repos, user preferences) into version chains where the same environment evolves across releases, and the agent must adapt to changes while preserving still-valid prior behavior. The paper also proposes EvoMem, a git-like memory layer that logs non-additive memory updates as append-only patches (storing before/after state, rationale, and triggering evidence) and retrieves relevant patches alongside the latest memory at inference.
Full breakdown →Toward Generalist Autonomous Research via Hypothesis-Tree Refinement
Across six real research tasks (model training, harness engineering, data synthesis) under a 48-hour budget and identical task interface, Arbor got the best held-out result on all six and roughly 2.5x the average relative held-out gain of Codex and Claude Code. On MLE-Bench Lite it reached 86.36% Any Medal with GPT-5.5, though this is a small task suite and baseline numbers are taken from an external leaderboard rather than re-run.
arXiv:2606.11926 · 126▲ · cs.CL, cs.AI
- Agents
- LLMs
Arbor is a framework for autonomous research (which the authors call Autonomous Optimization) where an LLM agent iteratively improves a code artifact against a fixed objective without step-level human supervision. Its core mechanism is Hypothesis Tree Refinement (HTR): a persistent tree whose nodes bind a hypothesis, a git-worktree artifact version, evaluation evidence, and a distilled insight, managed by a long-lived coordinator that dispatches short-lived executors and only promotes changes that pass a held-out test evaluator.
Full breakdown →OmniDirector: General Multi-Shot Camera Cloning without Cross-Paired Data
On a self-constructed 1,094-sample evaluation set, OmniDirector reached 72.74% translation precision (T-Pre) versus 52.21% for CamCloneMaster, a 39.3% relative gain, plus lower rotation/translation errors and much higher transition accuracy (96.52% temporal vs 38.94% for the only baseline that attempts transitions). Note the eval set and several metrics (leakage, semantic transition) are judged by the authors' own pipeline using Gemini 3.1 Pro, not an external benchmark.
arXiv:2606.13432 · 113▲ · cs.CV, cs.AI
- Vision
- Multimodal
OmniDirector is a video generation framework that clones camera motion from a reference video by rendering the extracted camera poses as a 'camera grid' (a video of grid lines moving through an empty 3D room), then feeding that grid into a Multi-Modal Diffusion Transformer via token concatenation. It also uses a hierarchical prompt expansion agent built on Qwen3-VL to describe camera motion (split into inter-shot and intra-shot descriptions) and fuse it with the reference image and user prompt at inference time.
Full breakdown →SpatialClaw: Rethinking Action Interface for Agentic Spatial Reasoning
59.9% average accuracy across 20 spatial reasoning benchmarks with a Gemma4-31B backbone, beating the structured tool-call agent SpaceTools by 11.2 points. In a controlled comparison holding the toolset fixed, SpatialClaw scored 59.9% vs 56.7% for structured tool-calls and 55.2% for single-pass code. Caveat: benchmarks over 1,000 samples were subsampled to 1,000, and the authors note remaining errors are dominated by perception quality rather than the interface, so further interface gains are limited.
arXiv:2606.13673 · 110▲ · cs.CV, cs.AI
- Agents
- Multimodal
- Vision
SpatialClaw is a training-free agent framework that gives a vision-language model a persistent Python kernel (preloaded with input frames, perception tools like SAM3 and Depth Anything 3, and libraries like NumPy/SciPy) and lets it write and run one code cell per step, inspecting intermediate results before writing the next cell. This 'code as the action interface' approach replaces two prior designs: single-pass code (commit to a full program before seeing any output) and structured JSON/XML tool-calls (fixed API menu).
Full breakdown →MaxProof: Scaling Mathematical Proof with Generative-Verifier RL and Population-Level Test-Time Scaling
With MaxProof scaling, the M3 model scored 35/42 on IMO 2025 and 36/42 on USAMO 2026, above the human gold-medal threshold on both. Caveat: these are tiny evals (6 problems each), scoring is done by the same model family's verifier plus stated expert judges, and the config is expensive (32 initial candidates, 4 verifier samples each, 10 refinement rounds); IMO 2025 Problem 6 still scored 0/7.
arXiv:2606.13473 · 92▲ · cs.LG, cs.AI, cs.CL
- LLMs
- RL
- Inference
MaxProof is a test-time scaling framework for competition math proofs that treats one merged model as generator, verifier, refiner, and ranker, then runs an evolution-style population search (dual PATCH/REWRITE refinement, conservative min-score fitness, pairwise tournament final selection). The underlying MiniMax-M3 model is trained through three specialist stages (proof generation via CISPO RL under a four-layer generative verifier, error-finding verification, and critique-conditioned repair) before being merged into a single release model.
Full breakdown →Subscribe
Get the next issue.
Free. One email a week. Unsubscribe any time: no account, no dark patterns.