--- title: AI Model Architectures url: "https://goremy.ai/blog/topic/ai-model-architectures" type: topic updated: "2026-11-06T00:00:11.390Z" posts: 26 --- # AI Model Architectures Explainers and tests of unconventional AI model designs: looped and recurrent transformers, diffusion LLMs, world models and decision models. Answers: How do looped and recurrent transformers differ from standard LLMs? ## Posts (26) - [How Is Ripple-65B Built? Dense Thin-Deep vs Mixture of Experts](https://goremy.ai/blog/articles/ripple-65b-thin-deep-dense-architecture-explained) (2026-11-06): Ripple-65B is a dense 80-layer transformer where all 65B parameters run per token. How it differs from MoE, plus its attention tweaks and training. - [Tencent Open-Source World Model vs Genie 4: Can You Run It?](https://goremy.ai/blog/articles/tencent-hy-world-model-open-source-genie-4-alternative) (2026-11-04): Tencent's open-weight world model claims Genie 4-style interactive generation on one consumer GPU, with minute-long keyboard steering. How it compares. - [What Is a Looped Transformer? NVIDIA Dory Explained](https://goremy.ai/blog/articles/what-is-looped-transformer-nvidia-dory-explained) (2026-11-03): A looped transformer reruns a shared block of layers per token to gain depth without more weights. Here is how NVIDIA's Dory does it, and what it costs. - [How Does SenseNova-Loop Get More Depth Without More Parameters?](https://goremy.ai/blog/articles/sensenova-loop-looped-transformer-mhc-explained) (2026-11-03): SenseNova-Loop-9B-Preview reuses 24 layers twice for 56 layer executions from 32 stored layers, with two-stream mHC and 3:1 linear-to-full attention. - [How Does SenseNova-Loop-9B-Preview's Looped Transformer Work?](https://goremy.ai/blog/articles/sensenova-loop-9b-preview-looped-transformer) (2026-11-03): SenseNova-Loop-9B-Preview reuses 24 layers twice with two-stream mHC and 3:1 hybrid attention: ~9B stored parameters, ~14.6B at inference, 256K context. - [How Is LFM3's Tri-Hybrid Architecture Built and Trained?](https://goremy.ai/blog/articles/lfm3-1-7b-base-training-recipe-architecture) (2026-11-02): How Liquid AI built LFM3-1.7B-Base: 13 gated short convolutions, 5 sliding-window and 6 global attention layers, and a 22T-token recipe to 256K context. - [What Is Clover-35B-A3B? Block Diffusion on a Qwen3.6 MoE](https://goremy.ai/blog/articles/clover-35b-a3b-block-diffusion-architecture-explained) (2026-10-30): How DiffuSpace's Clover derives from Qwen3.6-35B-A3B: bidirectional GDN attention, confidence-threshold decoding, and block-level expert coreset selection. - [Is NCM Faster Than an LLM? Next Concept Modeling Tested](https://goremy.ai/blog/articles/next-concept-modeling-ncm-stanford-speed-tests) (2026-10-29): Stanford's NCM writes text coarse to fine. Independent DGX Spark tests show about 4x speed for one user, slower batching, and a Sudoku result with a caveat. - [What Is Harness²? Self-Evolving Agent Harness Explained](https://goremy.ai/blog/articles/harness-squared-self-evolving-agent-harness-explained) (2026-10-27): Harness² lets an agent rewrite its skills, tools, sub-agents and control loop. How propose, probe, compose works and why a smaller model beat a stronger one. - [What Is DecisionNode? A Hands-On Look at Typed Decisions](https://goremy.ai/blog/articles/decisionnode-typed-decision-model-hands-on) (2026-10-26): DecisionNode returns typed answers with calibrated confidence instead of chat text. See tests on image boosters, API routing, the CLI, injection and RAG. - [Why Is Flint So Fast? MoE, Sliding Window, 4-Bit Output Layer](https://goremy.ai/blog/articles/flint-moe-architecture-explained-sliding-window-quantized-output) (2026-10-24): Flint runs about 10x faster than a dense 27B model. See how 2.5B active parameters, hybrid attention, a 4-bit output layer and a fused kernel do it. - [Matrix Game 4.0 vs Nvidia StreamRender H3: How Do They Render Games?](https://goremy.ai/blog/articles/matrix-game-4-0-vs-nvidia-streamrender-h3) (2026-10-23): Matrix Game 4.0 hands a video model a low-poly render. Nvidia StreamRender H3 hands it a color-coded map. Here is how the two approaches differ. - [Dory 1.0 4B Benchmarks: How Do 1, 2 and 4 Loops Compare?](https://goremy.ai/blog/articles/dory-4b-benchmarks-vs-qwen-3-5-gemma-4) (2026-10-22): NVIDIA Dory1.0-4B scores 50.3 to 61.8 on SWE-bench Verified from 1 to 4 loops. See how it compares with Qwen3.5-4B, Qwen3.5-9B and Gemma 4. - [Do Diffusion LLMs Scale? Mercury 3 vs Claude Haiku 5.5 and GPT-6 Luna](https://goremy.ai/blog/articles/do-diffusion-llms-scale-mercury-3-benchmarks) (2026-10-22): Mercury 3 is a diffusion LLM that hits 1,068 tokens/s. See how it differs from autoregressive models and how it compares with Haiku 5.5 and GPT-6 Luna. - [What Is Mercury Decide? Inception's Diffusion Decision Model Explained](https://goremy.ai/blog/articles/what-is-mercury-decide-diffusion-decision-model) (2026-10-21): Mercury Decide returns an answer and probability in about 0.16 seconds, now with images. How it works, its independent evals, and how to pair it with Mercury 3. - [What Is NVIDIA Nemotron Dory 1.0 4B? Looped Model Explained](https://goremy.ai/blog/articles/nvidia-nemotron-dory-1-0-4b-looped-model) (2026-10-21): Dory1.0-4B is NVIDIA's looped 4B language model that reuses a shared layer block to scale compute per token. Architecture, loop counts, and benchmarks. - [AI Decision Models vs. LLMs: Which Should You Use?](https://goremy.ai/blog/articles/ai-decision-models-vs-llms-which-should-you-use) (2026-10-21): AI decision models vs. LLMs compared on how they work, cost, latency, and benchmark results, with guidance on when to use each for routing and classification. - [AI Decision Models vs. Execution Models: Which Does What?](https://goremy.ai/blog/articles/ai-decision-models-vs-execution-models-which-does-what) (2026-10-20): Decision models return typed choices and scores; execution models and sandboxed code act on them. Compare latency, cost, and real benchmarks. - [CLM vs Jev: The System 1 Decision Model Comparison](https://goremy.ai/blog/articles/clm-vs-jev-system-1-decision-models-compared) (2026-10-04): Jev made fast, cheap decision models mainstream. CLM makes them self-hostable and fine-tunable. A numbers-first comparison of speed, cost, accuracy, and ownership. - [What Is RLCD AI? The Decision Architecture Behind Jev](https://goremy.ai/blog/articles/what-is-rlcd-ai-the-decision-architecture-behind-jev) (2026-09-28): RLCD trains models like Jev to output calibrated decisions, not text. Here's how it works, the real speed and cost numbers, and why 'can't hallucinate' isn't 'can't be wrong.' - [What Are Jev-Class Models? The Fast, Cheap Alternative to LLMs for Internal Tools](https://goremy.ai/blog/articles/what-are-jev-models-the-fast-cheap-llm-alternative) (2026-09-26): Jev-class decision models trade text generation for typed choices, scores, and probabilities, running 40-400x cheaper and up to 200x faster than chat LLMs. - [System 1 AI Models vs. JEV: What's the Difference?](https://goremy.ai/blog/articles/system-1-ai-model-vs-jev-the-difference-explained) (2026-09-24): What is a System 1 AI model versus JEV? A grounded look at fast, calibrated classifiers versus reasoning LLMs for GUI and form-filling agents. - [What Is an Infinite-Parameter LLM, and What Does It Change for Data Engineering?](https://goremy.ai/blog/articles/what-is-an-infinite-parameter-llm-a-plain-english-guide) (2026-09-19): Infinite-parameter LLMs turn live data into model weights instead of prompts. Here's how they work, and what they change for data engineering teams. - [What Are Looped Transformers in AI?](https://goremy.ai/blog/articles/what-are-looped-transformers-in-ai-explained) (2026-09-08): Looped transformers reuse the same layers repeatedly to reason in hidden math, not words. Here's how they work, why OpenAI's Astra uses them, and the safety tradeoff. - [What Is Recurrent Architecture AI? GPT-6 Astra's Looping Explained](https://goremy.ai/blog/articles/what-is-recurrent-architecture-ai-gpt-6-astra-explained) (2026-09-07): Recurrent depth loops shared transformer layers instead of stacking new ones. Here's how it powers GPT-6 Astra and why it raises governance concerns. - [What Is AI World Modeling?](https://goremy.ai/blog/articles/what-is-ai-world-modeling-a-plain-explainer) (2026-09-03): AI world modeling means predicting the next state of an environment, not just the next word. Here's what it is, where it came from, and why it matters for coding agents.