--- title: Build World Models for Interactive Simulation with... description: NVIDIA Cosmos Framework and Cosmos NIM let developers build and deploy world models for physical AI simulation. Cosmos Framework offers agent skills and... canonical: https://daily.dev/posts/build-world-models-for-interactive-simulation-with-nvidia-cosmos-ba6dexget twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Build World Models for Interactive Simulation with NVIDIA Cosmos | daily.dev og:description: NVIDIA Cosmos Framework and Cosmos NIM let developers build and deploy world models for physical AI simulation. Cosmos Framework offers agent skills and... og:url: https://daily.dev/posts/build-world-models-for-interactive-simulation-with-nvidia-cosmos-ba6dexget og:image: https://api.daily.dev/og/posts/ba6dEXgET.png og:image:alt: Build World Models for Interactive Simulation with NVIDIA Cosmos og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Build World Models for Interactive Simulation with NVIDIA Cosmos **[NVIDIA Developer](https://daily.dev/sources/nvidiadev)** · 10 min read · 0 upvotes · 0 comments ## Summary NVIDIA Cosmos Framework and Cosmos NIM let developers build and deploy world models for physical AI simulation. Cosmos Framework offers agent skills and recipes for training, post-training (including DPO, GRPO, PPO, OPD, and DiffusionNFT), and inference, extending beyond Cosmos 3 to Alpamayo, GR00T, Qwen, and pi0.5. New Cosmos 3 Nano simulation recipes and example checkpoints support autonomous vehicle camera and LiDAR generation, robot simulation, and surgical simulation. Cosmos NIM adds optimized containers with quantization and speculative decoding, with speculative decoding delivering roughly 2 to 3 times mean speedups on H200 GPUs and streaming VLM inference enabling sub-second alerts. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much speedup does speculative decoding give for NVIDIA Cosmos NIM inference on H200 GPUs? Speculative decoding delivers measured mean speedups of roughly 2 to 3 times on H200 GPUs for Cosmos NIM inference. This is combined with quantization, caching, token pruning, and KV-cache-aware routing to further reduce memory use and latency, alongside sub-second alerting with 10x lower latency for streaming video understanding use cases. _daily.dev surfaces benchmark details like this for engineers evaluating GPU inference performance options._ ### How much GPU VRAM does Cosmos3-Nano need for generation tasks in BF16 versus FP8? Cosmos3-Nano generation requires 58 GiB of VRAM in BF16, 44 GiB in FP8, and 31 GiB in FP8 with offloading. For reasoning tasks, Cosmos3-Nano needs 23.1 GiB in both BF16 and FP8. Cosmos3-Super generation needs 150 GiB BF16, 93 GiB FP8, or 38 GiB with FP8 plus offloading. _teams sizing GPU budgets for world-model deployments track specs like these on daily.dev._ ### What is the difference between Cosmos3-Nano-Transfer-Auto and Cosmos3-Nano-Sim-Auto for autonomous vehicle simulation? Cosmos3-Nano-Transfer-Auto uses bidirectional generation to denoise an entire multi-view clip at once for offline synthetic data generation, supporting up to 400 frames at 480p or 720p. Cosmos3-Nano-Sim-Auto is its causal, few-step-distilled variant enabling autoregressive streaming generation, producing observations incrementally as a driving policy updates the scenario for closed-loop evaluation. _developers choosing between offline data generation and closed-loop AV simulation can compare these approaches on daily.dev._ ## Similar posts on daily.dev - [NVIDIA Cosmos Opens New Tools to Train, Evaluate and Deploy World Models](https://daily.dev/posts/nvidia-cosmos-opens-new-tools-to-train-evaluate-and-deploy-world-models-i61wzilhw) · NVIDIA · 0 upvotes · 0 comments - [How to Scale Data Generation for Physical AI with the NVIDIA Cosmos Cookbook](https://daily.dev/posts/how-to-scale-data-generation-for-physical-ai-with-the-nvidia-cosmos-cookbook-5wuqobudk) · NVIDIA Developer · 0 upvotes · 0 comments - [Introducing NVIDIA Cosmos Policy for Advanced Robot Control](https://daily.dev/posts/introducing-nvidia-cosmos-policy-for-advanced-robot-control-kq1bpwgka) · Hugging Face · 0 upvotes · 0 comments --- Tags: [#world-models](https://daily.dev/tags/world-models) [View this post on daily.dev](https://daily.dev/posts/build-world-models-for-interactive-simulation-with-nvidia-cosmos-ba6dexget) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Build World Models for Interactive Simulation with NVIDIA Cosmos","url":"https://daily.dev/posts/build-world-models-for-interactive-simulation-with-nvidia-cosmos-ba6dexget","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/build-world-models-for-interactive-simulation-with-nvidia-cosmos-ba6dexget"},"datePublished":"2026-11-02T17:05:07.330Z","dateModified":"2026-11-02T17:12:02.201Z","description":"NVIDIA Cosmos Framework and Cosmos NIM let developers build and deploy world models for physical AI simulation. Cosmos Framework offers agent skills and...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/cc9b4688ace066d95a84de3f6093d9ee?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/cc9b4688ace066d95a84de3f6093d9ee?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"NVIDIA Developer","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"NVIDIA Developer","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/86e45aab42ba48ce83103d01b1119910","url":"https://daily.dev/sources/nvidiadev"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/build-world-models-for-interactive-simulation-with-nvidia-cosmos-ba6dexget","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"world-models","timeRequired":"PT10M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"NVIDIA Developer","item":"https://daily.dev/sources/nvidiadev"},{"@type":"ListItem","position":3,"name":"Build World Models for Interactive Simulation with NVIDIA Cosmos"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/build-world-models-for-interactive-simulation-with-nvidia-cosmos-ba6dexget#faq","mainEntity":[{"@type":"Question","name":"How much speedup does speculative decoding give for NVIDIA Cosmos NIM inference on H200 GPUs?","acceptedAnswer":{"@type":"Answer","text":"Speculative decoding delivers measured mean speedups of roughly 2 to 3 times on H200 GPUs for Cosmos NIM inference. This is combined with quantization, caching, token pruning, and KV-cache-aware routing to further reduce memory use and latency, alongside sub-second alerting with 10x lower latency for streaming video understanding use cases. daily.dev surfaces benchmark details like this for engineers evaluating GPU inference performance options."}},{"@type":"Question","name":"How much GPU VRAM does Cosmos3-Nano need for generation tasks in BF16 versus FP8?","acceptedAnswer":{"@type":"Answer","text":"Cosmos3-Nano generation requires 58 GiB of VRAM in BF16, 44 GiB in FP8, and 31 GiB in FP8 with offloading. For reasoning tasks, Cosmos3-Nano needs 23.1 GiB in both BF16 and FP8. Cosmos3-Super generation needs 150 GiB BF16, 93 GiB FP8, or 38 GiB with FP8 plus offloading. teams sizing GPU budgets for world-model deployments track specs like these on daily.dev."}},{"@type":"Question","name":"What is the difference between Cosmos3-Nano-Transfer-Auto and Cosmos3-Nano-Sim-Auto for autonomous vehicle simulation?","acceptedAnswer":{"@type":"Answer","text":"Cosmos3-Nano-Transfer-Auto uses bidirectional generation to denoise an entire multi-view clip at once for offline synthetic data generation, supporting up to 400 frames at 480p or 720p. Cosmos3-Nano-Sim-Auto is its causal, few-step-distilled variant enabling autoregressive streaming generation, producing observations incrementally as a driving policy updates the scenario for closed-loop evaluation. developers choosing between offline data generation and closed-loop AV simulation can compare these approaches on daily.dev."}}]} ```