groundy

models & research

103 articles·rss

Top in models & research


  1. jul 24modelsDiffusion LLMs: Training Cost, Not Parallel Decoding, Drives Deployment
  2. jul 24modelsOpen-Weight Routers vs Fable 5: The Routing Math That Actually Matters
  3. jul 23modelsDeepSeek-V4 1M Context vs RAG: Why Retrieval Stays
  4. jul 23modelsQwen-Image-3.0 Does Not Exist: Why Self-Hosting Image Models Is Premature
  5. jul 20modelsKimi K3: 2.8T Parameters, MoE Routing, and Self-Hosting Reality
  6. jul 20modelsKimi K3 vs Qwen3.8 Max: Routing Strategy for July 2026
  7. jul 20modelsQwen3.8 Max Preview: Missing Benchmarks, Weights, and Pricing
  8. jul 19modelsHuggingFace 100x Inference: Generalizable vs Platform-Locked Optimizations
  9. jul 19modelsKimi K3 Code Arena Rank: Self-Hosting Cost Math for Coding Agents
  10. jul 14modelsCan Tool-Adaptive LLM Rerankers Improve RAG Without Always Calling Tools?
  11. jul 14modelsDoes Speculative Decoding with Progressive Tree Drafting Cut LLM Latency?
  12. jul 10modelsAnalytic Inference Cuts Bayesian Deep Ensemble Serving Cost, But Leaves Training as the Bottleneck
  13. jul 10modelsTree-of-Thoughts Improves Text-to-Image Prompting by Reasoning Over Hypotheses, Not Pixels
  14. jul 10modelsFourierQK's spectral Q/K filter cuts TinyShakespeare loss by 79%, but long-context proof is missing
  15. jul 10modelsTencent Hunyuan 3's Agent Push Has No Public DeepSeek or Qwen Benchmarks Yet
  16. jul 10modelsWhen Does Memory, Not Compute, Decide Who Can Profitably Serve LLMs?
  17. jul 10modelsCan We Trust LLM Logic? A Graph-Based Stress Test Finds Three Failure Modes
  18. jul 09modelsLongCat-2.0 Hits Claude Opus 4.6 Class on Agents From a 50K-GPU Cluster
  19. jul 08modelsWhen Do Time Series Foundation Models Pay Off? The Break-Even Threshold
  20. jul 07modelsVLA Grounder Tests Language Conditioning to Optimize Black-Box Vision-Action Models
  21. jul 07modelsBlack-Box LLM Architecture Inference: What API Restrictions Reveal About Hidden Model Structure
  22. jul 07modelsInduceKV Tests Continual Learning for Multimodal LLMs Without Expanding Cache
  23. jul 03modelsSonnet 5 vs GPT-5.5: Pricing, Benchmarks, and the Switching Math
  24. jun 30modelsHow LLMs Fuse Conflicting Facts: Single-Source vs Multi-Source Truth
  25. jun 29modelsLinear Transformers Get a Learnable Kernel: Does Flexformer Change the Efficiency Tradeoff?
  26. jun 29modelsHuawei Ships CUDA-Free AI Compute On-Device, but Ascend Quantization Accuracy Is Unverified
  27. jun 29modelsDo Multimodal RAG Models Ignore Late Evidence? A Primacy Bias Test
  28. jun 29modelsCan Deep Learning Design RF Power Amplifiers Without Full EM Simulation?
  29. jun 28modelsSynthetic Clinical Notes from LLMs: Believable Prose Is Not Clinical Validity
  30. jun 28modelsDoubao vs Qwen 3.7 vs GLM-5.2: Route by Axis, Not Leaderboard
  31. jun 28modelsCan Dynamic Experts Fix Catastrophic Forgetting in Robot Manipulation?
  32. jun 28modelsError-Conditioned Neural Solvers vs Iterative Refinement: When Does Learned Correction Win?
  33. jun 28modelsVision-Language Models Move Past Object Detection: The MLLM Perception Shift
  34. jun 28modelsCan Autoregressive Boltzmann Generators Replace MCMC in Simulation?
  35. jun 28modelsLook-Before-Move Plans Observation Before Motion in Dynamic 3D Story Worlds
  36. jun 27modelsGLM 5.2, Qwen 3.7, and DeepSeek in 2026: A Routing Map by Workload, Not by Rank
  37. jun 27modelsSLM Pipeline Catches 10% of Papers Human Reviewers Missed, but No Model Matched Human Accuracy
  38. jun 27modelsMiniMax M3 vs GLM-5.2: Whose 1M-Context Claim Holds Up?
  39. jun 27modelsDeepSeek V4.1 Flash vs Qwen 3.7 vs Llama 4.5: June 2026 HF Trending Ranks Velocity, Not Installs
  40. jun 27models125 Targeted Wikipedia Edits Left a Detectable Signal in Llama Pretraining
  41. jun 27modelsCan SAE Features Stop LLMs From Forgetting During Continual Learning?
  42. jun 27modelsCan a 30B Model Post-Train Itself? A-Evolve-Training Tests Autonomous RL
  43. jun 26modelsOpen-Weight LLM Leaderboards 2026: Where DeepSeek, Qwen, and GLM Rank
  44. jun 26modelsQwen3.7-Max's Top-Ranked Claim vs the Artificial Analysis Index
  45. jun 26modelsDoes Tree-of-Thought Reasoning Scale to Billion-User Modeling?
  46. jun 26modelsCan LLMs Debug Verilog? VeriPilot Puts an Agent on RTL Errors
  47. jun 25modelsTask Decomposition Helps LLMs by Shrinking Output Space, Not by Cutting Labeling Cost
  48. jun 25modelsFlow Matching vs U-Net: A Skip-Free Backbone for Speech Models
  49. jun 25modelsA Per-Neuron Sequence Model Was Withdrawn From arXiv as Coverage Hailed It
  50. jun 25modelsPV-TAM Corrects Decoding Drift and Boundary-Marker Bias in VLM Localization Scoring

Every new foundation model arrives wrapped in a benchmark chart and a press release. This beat exists because the chart almost never tells you what the model actually does — and the press release tells you even less. The interesting work is one layer down: the attention variant that changes the cost curve, the parameterization fix that makes scaling laws actually transfer, the eval protocol whose failure modes flip the standings once you change a decoding parameter or a codec. We cover that layer.

We treat open-weight and closed releases as the same story told from opposite ends of a distribution shift. A trillion-parameter mixture beating a dense competitor on one harness and losing on another is not a contradiction; it is evidence about what the harness measures. Training-efficiency claims, context-window claims, and reasoning-benchmark claims all get the same treatment — read the method section, find the assumption that was load-bearing, and report whether removing it changes the result. When a paper proposes a new safety mechanism, a new compression trick, or a new continual-learning split, we are interested in the part the authors did not want to highlight.

The throughline is comparative and skeptical without being contrarian. Foundation-model research has become the field where the largest gap between published numbers and deployed behavior tends to live. Closing that gap — with reproduction notes, ablation reading, and honest accounting of what generalizes versus what was tuned into the eval — is the beat.