← the graph
Sector 8 · 11 sub-themes

Open Models & Ecosystem

DeepSeek's silent drop of V4-Pro 0813 is the headline: a 1.6T-parameter MoE model that scores 87.9 on Terminal Bench — a 15.8% gain over its April Preview — while pricing at roughly 57x cheaper than Claude Fable 5. No announcement, just an overnight GA flip. That pricing-to-performance ratio is what makes it disruptive, not just competitive. NVIDIA is pressing hard on multiple fronts simultaneously: Nemotron 3.5 Lightning (30B MoE, 3B active at inference, one-third the cost of Opus 4.8) targets always-on agents, Alpamayo 2 Super opens autonomous-driving frontier reasoning to automakers, and a rumored Nemotron 4 at trillion-plus parameters signals NVIDIA's intent to challenge the best closed models outright. Alibaba delivered on its promise with Qwen3.8-27B under Apache 2.0, though the 2.4T-parameter Max variant remains closed. Mistral pivoted from pure first-party host to open-model marketplace, onboarding Zhipu AI's GLM-5.2 — a move that signals European open infrastructure potentially becoming neutral ground for Chinese labs. Kimi K3 landing in GitHub Copilot via Fireworks AI shows open-weight models are now first-class citizens in enterprise dev tooling. Unsloth Desktop is quietly significant: the first single app combining local inference and fine-tuning across Mac, Windows, and Linux, collapsing a workflow that previously required separate toolchains. The ecosystem is fragmenting upward in scale and downward to the desktop simultaneously.

DeepSeekNVIDIAAlibabaMistralMoonshot AIZhipu AIFireworks AIUnslothV4-Pro 0813Nemotron 3.5 LightningAlpamayo 2 SuperQwen3.8-27Bopen-weight model releasesmixture-of-experts scalingpricing-to-performance disruptionlocal inference and fine-tuningautonomous driving AIenterprise developer toolingopen-model marketplacebenchmark performance gainsedge-to-hyperscale fragmentationChinese AI lab competition
8.1

Alibaba Releases Qwen3.8-27B Open Weights; Qualcomm-Hugging Face Partnership Expands

  • Alibaba's Qwen team released open weights for Qwen3.8-27B, a 27-billion-parameter native multimodal dense model, published on Hugging Face under the Apache 2.0 license, fulfilling a previously made promise to release open weights for the Qwen3.8 series. [6][7][8][9]
  • The Qwen3.8-27B model is positioned for local deployment, with the Apache 2.0 license enabling broad commercial and research use without restrictions. [8][9]
  • Questions remain about the larger Qwen3.8-Max model, described as having 2.4 trillion parameters, whose weights have not been released as of the reporting. [10]
  • Qualcomm and Hugging Face expanded their strategic partnership to advance open AI development across edge and cloud environments, with a focus on onboarding AI models across both device and cloud deployments. [1][2][3][4][5]
AlibabaQwen teamQwen3.8-27BHugging FaceQwen3.8-MaxQualcommApache 2.0open weights releasemultimodal dense modellocal deploymentedge AIcloud deploymentstrategic partnershipcommercial licensingQwen model series
Sources
  1. [1]Qualcomm — Qualcomm and Hugging Face Expand Relationship to Advance Open, Developer-Driven AI from Device to Cloud★news
  2. [2]Crypto Briefing — Qualcomm expands strategic relationship with Hugging Face to onboard AI models across edge and cloud★news
  3. [3]Yahoo Finance — Qualcomm (QCOM), Hugging Face Expand Partnership to Accelerate Open AI Development★news
  4. [4]The Fast Mode — Qualcomm, Hugging Face Expand Partnership to Advance Open AI from Edge to Cloud★news
  5. [5]GuruFocus — Qualcomm (QCOM) Expands Partnership with Hugging Face for AI Dev★news
  6. [6]Alibaba_Qwen — We promised open weights for Qwen3.8. Now, time to meet them! 🎉 ⚡ Qwen3.8-27B: - A native multimodal dense model. With just 27B parameters, ★twitter
  7. [7]Alibaba_Qwen — We promised open weights for Qwen3.8. Now, time to meet them! 🎉 ⚡ Qwen3.8-27B: - A native multimodal dense model. With just 27B parameters, ★twitter
  8. [8]testingcatalog — Alibaba published Qwen3.8-27B weights on Huggingface under the Apache 2.0 license! > Qwen3.8-27B is a native multimodal dense model that you★twitter
  9. [9]rohanpaul_ai — Alibaba dropped the weights for Qwen3.8-27B as a 27B open-weight multimodal model built for local deployment. - Apache 2.0 weights and suppotwitter
  10. [10]TeksEdge — 🔥 THIS is the Qwen3.8 model Local AI users have been waiting for! Questions still remain! The 2.4 TRILLION parameter Qwen3.8-Max weights wertwitter
8.2

NVIDIA open-sources Alpamayo 2 Super frontier autonomous driving model

  • NVIDIA launched Alpamayo 2 Super, a frontier open reasoning model with 34 billion parameters designed for autonomous vehicles; described as going beyond perception to understanding, the model is released with open weights allowing automakers to fine-tune it on their own fleet data while retaining full ownership of that data. [2][3][4][5][6][7]
  • NVIDIA separately open-sourced DFlash, a block diffusion model that accelerates autoregressive LLM inference by up to 15× and integrates with vLLM. [1]
NVIDIAAlpamayo 2 SuperDFlashvLLMopen reasoning modelautonomous vehiclesfrontier modelopen weightsfleet data fine-tuningblock diffusion modelLLM inference accelerationautoregressive inference
Sources
  1. [1]Open Source For You — NVIDIA has open-sourced DFlash, a block diffusion model that speeds up autoregressive LLM inference by up to 15× while integrating with vLLM★news
  2. [2]JensenHuang — Today, we’re launching Alpamayo 2 Super, our frontier open reasoning model for autonomous vehicles. Beyond seeing, Alpamayo understands and ★twitter
  3. [3]JensenHuang — Today, we’re launching Alpamayo 2 Super, our frontier open reasoning model for autonomous vehicles. Beyond seeing, Alpamayo understands and ★twitter
  4. [4]JensenHuang — Today, we’re launching Alpamayo 2 Super, our frontier open reasoning model for autonomous vehicles. Beyond seeing, Alpamayo understands and ★twitter
  5. [5]JensenHuang — Today, we’re launching Alpamayo 2 Super, our frontier open reasoning model for autonomous vehicles. Beyond seeing, Alpamayo understands and ★twitter
  6. [6]JensenHuang — Today, we’re launching Alpamayo 2 Super, our frontier open reasoning model for autonomous vehicles. Beyond seeing, Alpamayo understands and ★twitter
  7. [7]shawnchauhan1 — Automakers can now fine-tune a frontier driving model and keep every mile of fleet data private. NVIDIA open-sourced Alpamayo 2 Super, a reatwitter
8.3

Unsloth Desktop launches open-source local AI training and inference app

  • Unsloth, a YC-backed company, launched "Unsloth Desktop," described as the first desktop app to support both running and fine-tuning/training AI models locally within a single application; it is open-source and runs on Mac, Windows, and Linux. [1][3]
  • The app supports image, video, and audio models in addition to text, distinguishing it from LM Studio, which one observer described it as a "definitive replacement" for given that LM Studio lacks local training capability. [4]
  • NVIDIA's RTX Spark account highlighted Unsloth Desktop specifically for combining training and inference in one local desktop app, underscoring the novelty of that dual capability. [2]
UnslothUnsloth DesktopY CombinatorLM StudioNVIDIARTX Sparklocal AI desktop appon-device fine-tuninglocal model traininginference and training combinedopen-source AI toolsmultimodal model supportconsumer hardware AI
Sources
  1. [1]UnslothAI — Introducing Unsloth Desktop 🦥 The first desktop app to run and train models locally. • Open-source. Runs on Mac, Windows and Linux • Support★twitter
  2. [2]NVIDIARTXSpark — Training AND running AI models locally from the same desktop app? Yes please. 🔥 Super cool to see @UnslothAI Desktop bringing training, fine★twitter
  3. [3]UnslothAI — Introducing Unsloth Desktop 🦥 The first desktop app to run and train models locally. • Open-source. Runs on Mac, Windows and Linux • Supporttwitter
  4. [4]pigeon__s — Seems like a definitive LM Studio replacement. They're basically the same thing except Unsloth's: - supports image, video, and audio models,twitter
8.4

Mistral AI opens platform to third-party open models, starting with GLM-5.2

  • Mistral AI announced it is opening its platform to host third-party open-source models, running them on the same infrastructure as its own models, with the first integration being GLM-5.2 from Zhipu AI (Z.ai), a Beijing-based Chinese AI lab. [5][6][7]
  • GLM-5.2, developed by Zhipu AI, is an open-source model that has drawn attention in U.S. tech circles and has been benchmarked as challenging Anthropic and OpenAI in overall performance. [1][3][4]
  • Mistral's move to support third-party open models represents a strategic expansion from a purely first-party model host toward a broader AI model platform or marketplace. [2][5]
Mistral AIGLM-5.2Zhipu AIAnthropicOpenAIopen-source model hostingthird-party model integrationAI model marketplaceplatform expansionChinese AI developmentcompetitive benchmarkingfirst-party vs third-party models
Sources
  1. [1]Counterpoint Research — GLM-5.2 Challenges Anthropic and OpenAI in Overall Performance★news
  2. [2]The Tech Buzz — Mistral AI Takes on OpenAI with Open Source Frontier Models★news
  3. [3]Business Insider — What is GLM-5.2? Another open-source Chinese AI model has Silicon Valley's attention.★news
  4. [4]The Express Tribune — Chinese open-source AI model GLM-5.2 creates buzz in US tech circles★news
  5. [5]Techmeme — Mistral says its platform will support third-party open models, starting with Z․ai's GLM-5.2, and run them on the same infrastructure as itstwitter
  6. [6]etnshow — BREAKING: Mistral to begin hosting third-party open models on its platform, starting with GLM-5.2 from Zai, the Chinese AI lab, per Venture twitter
  7. [7]buildwithhassan — mistral is opening its platform to third-party open models, starting with https://t.co/L9Wrtjlp7i's GLM-5.2 it runs on mistral's own infrasttwitter
8.5

NVIDIA Releases Nemotron 3.5 Lightning Open 30B MoE Model

  • NVIDIA released Nemotron 3.5 Lightning, an open-weights 30B parameter Mixture-of-Experts model that activates only 3B parameters at inference time. The model employs a hybrid architecture with interleaved Mamba-2 and MoE layers alongside select Attention layers, and is released alongside speculative decoding methods (including DSpark speculative decoding tuned for DGX Spark) for faster text generation. [5][6][7][8][9]
  • Nemotron 3.5 Lightning is compatible with vLLM, SGLang, and Docker Model Runner, and is governed by the OpenMDW-1.1 model license. NVIDIA also launched NeMo Switchyard alongside the model. [9][10]
  • According to one source, Nemotron 3.5 Lightning cuts task completion costs to nearly one-third of Opus 4.8, and delivers 4× the output speed of similarly-sized models, positioning it for always-on agent workloads. [6][10]
  • Qualcomm and Hugging Face announced an expanded strategic relationship aimed at advancing open, developer-driven AI across edge devices and cloud environments, with a focus on onboarding AI models across both deployment tiers. [2][3][4]
  • NVIDIA open-sourced DFlash, a block diffusion model designed to accelerate autoregressive LLM inference by up to 15×, with integration support for vLLM. [1]
NVIDIANemotron 3.5 LightningNeMo SwitchyardQualcommHugging FaceDFlashvLLMSGLangDocker Model RunnerDSparkOpenMDW-1.1Opus 4.8mixture-of-experts LLM inference efficiencyhybrid Mamba-attention architecturespeculative decoding accelerationopen-weights model releaseedge and cloud AI deploymentblock diffusion autoregressive inferencealways-on AI agent workloadsdeveloper-driven open AI ecosystem
Sources
  1. [1]Open Source For You — NVIDIA has open-sourced DFlash, a block diffusion model that speeds up autoregressive LLM inference by up to 15× while integrating with vLLM★news
  2. [2]Qualcomm — Qualcomm and Hugging Face Expand Relationship to Advance Open, Developer-Driven AI from Device to Cloud★news
  3. [3]Crypto Briefing — Qualcomm expands strategic relationship with Hugging Face to onboard AI models across edge and cloud★news
  4. [4]The Fast Mode — Qualcomm, Hugging Face Expand Partnership to Advance Open AI from Edge to Cloud★news
  5. [5]NVIDIAAI — Introducing NVIDIA Nemotron 3.5 Lightning⚡ An open 30B MoE model with 3B active parameters, built for always-on agents to complete high-volu★twitter
  6. [6]RoundtableSpace — NVIDIA just dropped Nemotron 3.5 Lightning - an open 30B MoE model with only 3B active parameters. 4x the output speed of similar-sized mode★twitter
  7. [7]NVIDIAAI — Introducing NVIDIA Nemotron 3.5 Lightning⚡ An open 30B MoE model with 3B active parameters, built for always-on agents to complete high-volu★twitter
  8. [8]ArtificialAnlys — NVIDIA has just released the first Nemotron 3.5 model: Nemotron 3.5 Lightning, a highly efficient small open weights model with performance ★twitter
  9. [9]BrianRoemmele — Really love open source NVIDIA Nemotron 3.5 Lightning A 30B MoE model with 3B active parameters, built for always-on agents. It allows for htwitter
  10. [10]ai_for_success — NVIDIA just launched Nemotron 3.5 Lightning and NeMo Switchyard. Nemotron 3.5 Lightning cuts task completion costs to nearly 1/3 of Opus 4.8twitter
8.6

Moonshot AI's Kimi K3 open-weight model launches in GitHub Copilot

  • GitHub has made Kimi K3, an open-weight model from Moonshot AI, generally available in GitHub Copilot across Pro, Pro+, Max, Business, and Enterprise plans, hosted via Fireworks AI. The model is described as delivering frontier-level agentic coding abilities with cost-effective pricing of $3 per 1M input tokens, $15 per 1M output tokens, and $0.30 per 1M cached input tokens under usage-based billing. [1][2][3]
  • Rollout is gradual and the model is off by default for Business and Enterprise plans; administrators must explicitly enable a Kimi K3 policy in Copilot settings before any organization member can select it, and GitHub recommends reviewing open-weight models against security, compliance, and data-governance requirements before doing so. [1][2][3]
  • GitHub temporarily paused the Kimi K3 rollout to mitigate an incident with GitHub Actions, then resumed it, with billing switched to provider list pricing under usage-based billing. [1][2][3]
GitHubKimi K3Moonshot AIGitHub CopilotFireworks AIopen-weight model availabilityagentic codingGitHub Copilot integrationusage-based billingenterprise AI governancemodel access policycoding assistant deployment
Sources
  1. [1]chomado — 見逃してた、先週の公式記事 Kimi K3 が GitHub Copilot で使えます👀 https://t.co/Mh6IxEmb84twitter
  2. [2]chomado — 見逃してた、先週の公式記事 Kimi K3 が GitHub Copilot で使えます👀 https://t.co/Mh6IxEmb84twitter
  3. [3]Kesin11 — GitHub Copilotでも話題のKimi K3が利用可能に。Fireworks AIでホスティングされているとのこと。例によってBuisiness, Enterpriseはadminによるポリシー許可が必要 / “Kimi K3 is now available in Gitwitter
8.7

Google DeepMind Open-Sources WeatherNext Cyclones AI Forecasting System

  • Google DeepMind and Google Research jointly developed WeatherNext Cyclones, an AI model for predicting tropical cyclone track, intensity, and wind radius, publishing operational results in Nature and open-sourcing both the code and model weights. [2][3]
  • The WeatherNext Cyclones system operates at 0.25° global resolution and can simultaneously predict both cyclone trajectory and intensity — a capability that conventional models handle separately — extending warning lead time by approximately 30 hours compared to existing approaches. [4]
  • Google DeepMind and Google Research also launched Weather Lab, an interactive platform hosting AI-based cyclone prediction capabilities built on similar underlying technology. [1]
Google DeepMindGoogle ResearchWeatherNext CyclonesWeather LabNaturetropical cyclone predictionAI weather modelingtrack and intensity forecastingextended warning lead timeopen-source model weightsglobal resolution forecastingoperational meteorology
Sources
  1. [1]GoogleDeepMind — We’re using AI to improve cyclone prediction. 🌀 Introducing Weather Lab: a new interactive platform developed with @GoogleResearch, hosting ★twitter
  2. [2]LangChainJP — 【Google DeepMindの台風予報AI、運用結果をNatureに報告・コードとモデル重みを公開】 Google DeepMindとGoogle Researchは、熱帯低気圧（台風やハリケーン）の進路・強度・風域を予測するAIモデル「WeatherNext Cyclonetwitter
  3. [3]LangChainJP — 【Google DeepMindの台風予報AI、運用結果をNatureに報告・コードとモデル重みを公開】 Google DeepMindとGoogle Researchは、熱帯低気圧（台風やハリケーン）の進路・強度・風域を予測するAIモデル「WeatherNext Cyclonetwitter
  4. [4]DailyAIWireNews — https://t.co/ulsYqgMY9z Google DeepMind AI Enhances Cyclone Forecasting by 30 Hours Why it matters: Longer, more accurate cyclone warnings gtwitter
8.8

Nvidia Develops Nemotron 4 Trillion-Parameter Open-Source AI Model

  • Nvidia is developing Nemotron 4, a next-generation open-source AI model family whose largest variant is expected to exceed 1 trillion parameters, which would make it one of the largest openly released models to date. [2][3][4][5][6][7]
  • Nemotron 4 is positioned as Nvidia's competitive response to leading frontier models in both the open-source and closed-source categories, aiming to challenge the best open models in the world. [2][3][4][6][7]
  • Mistral AI is also taking on OpenAI by releasing open-source frontier models, indicating broader competitive pressure in the open-model ecosystem beyond Nvidia's entry. [1]
NvidiaNemotron 4Mistral AIOpenAIopen-source AI modelslarge language model competitiontrillion-parameter modelsfrontier model developmentopen vs closed source AINvidia AI strategy
Sources
  1. [1]The Tech Buzz — Mistral AI Takes on OpenAI with Open Source Frontier Models★news
  2. [2]StockSavvyShay — $NVDA is developing Nemotron 4 which is a 1T+ parameter open-source model aimed at competing with the best open models in the world. By push★twitter
  3. [3]wallstengine — Nvidia is developing Nemotron 4, a new open-source AI model expected to have at least 1 trillion parameters. $NVDA is aiming for it to compe★twitter
  4. [4]MTSlive — SITUATION DETECTED: Nvidia is developing Nemotron 4, a next generation open-source family of models to challenge the best open models in the★twitter
  5. [5]Cointelegraph — 🔥 LATEST: Nvidia is developing Nemotron 4, a new generation of open-source AI models to compete with leading open-source systems. https://t.★twitter
  6. [6]Sam_Badawi — Nvidia is going BIG on open-source AI with Nemotron 4. $NVDA is reportedly developing a 1T+ parameter model designed to compete with the wortwitter
  7. [7]crypto_banter — 🚨NVIDIA BUILDING TRILLION-PARAMETER OPEN-SOURCE AI MODEL! NVIDIA is developing a new Nemotron 4 open-source AI model, according to The Infortwitter
8.9

New Open Embedding Models Released Alongside Reasoning-Augmented Retrieval Research

  • Google DeepMind released **EmbeddingGemma**, described as a best-in-class open embedding model designed for on-device AI, at 308M parameters. The model covers 50+ languages within a single shared embedding space, enabling cross-lingual retrieval (e.g., searching in one language and retrieving documents written in another). [1][2]
  • IBM released **granite-embedding-small-english-r2**, a compact model designed to convert text into searchable vectors for feature extraction and semantic search tasks. [3]
  • A research paper introduced **GEM (Generative Embedding Model)**, which addresses the gap between how users express complex information needs and how conventional retrievers interpret surface-level query-document matches. GEM unifies generation and embedding in a single model by first reasoning over a query to determine user intent and relevance criteria, then appending an embedding token to encode the enriched context for retrieval. On reasoning-intensive and instruction-following retrieval benchmarks, GEM outperforms its non-reasoning variant and matches baselines that use substantially larger models; its generative design also enables test-time compute scaling via prompting to further boost retrieval performance. [4][5][6][7]
Google DeepMindEmbeddingGemmaIBMgranite-embedding-small-english-r2GEMGenerative Embedding Modelembedding modelssemantic searchon-device AIcross-lingual retrievalgenerative embeddingsreasoning-intensive retrievalinstruction-following retrievaltest-time compute scalingtext vectorizationmultilingual embeddings
Sources
  1. [1]GoogleDeepMind — EmbeddingGemma is our new best-in-class open embedding model designed for on-device AI. 📱 At just 308M parameters, it delivers state-of-the-★twitter
  2. [2]beaconfindapp — Search in one language, find the document written in another. The embedding model covers 50+ languages in a single shared space, so the word★twitter
  3. [3]HuggingModels — Ever needed a model that turns text into smart, searchable vectors? Meet granite-embedding-small-english-r2, a compact yet powerful feature-twitter
  4. [4]_reachsumit — GEM: A Generative Embedding Model Bridging Reasoning and Retrieval Presents a unified model that reasons about a query's intent before embedtwitter
  5. [5]_reachsumit — GEM: A Generative Embedding Model Bridging Reasoning and Retrieval Presents a unified model that reasons about a query's intent before embedtwitter
  6. [6]_reachsumit — GEM: A Generative Embedding Model Bridging Reasoning and Retrieval Presents a unified model that reasons about a query's intent before embedtwitter
  7. [7]_reachsumit — GEM: A Generative Embedding Model Bridging Reasoning and Retrieval Presents a unified model that reasons about a query's intent before embedtwitter
8.10

DeepSeek silently releases V4-Pro 0813 with major benchmark gains

  • DeepSeek silently released DeepSeek V4-Pro 0813 (a GA build dated August 13), a 1.6T-parameter MoE model with 49B active parameters and 1M context window, scoring 87.9 on Terminal Bench — a 15.8% improvement over their April Preview model — and seeing its DeepSWE agent score leap from 12.8 to 62.7. [2][4][6]
  • The model is priced at $0.435/M input tokens and $0.87/M output tokens, which sources describe as approximately 57x cheaper than Claude Fable 5 while approaching or matching it on several benchmarks. [3][5][1]
  • The deployment was described as an overnight silent flip to the 0813 GA build, with no formal announcement, consistent with DeepSeek's pattern of quiet releases. [2][6]
DeepSeekDeepSeek V4-Pro 0813Claude Fable 5Terminal BenchDeepSWEMoE large language modelsilent model releasebenchmark performancecost comparisonAI agent scoringcontext windowmodel pricing
Sources
  1. [1]cline — DeepSeek silently released V4-Pro 0813, up 15.8% on Terminal Bench from their April Preview model, with Fable 5 performance at ~57x cheaper ★twitter
  2. [2]PhilShteuck — DeepSeek just flipped deepseek-v4-pro to the 0813 GA build overnight. 1.6T total, 49B active, 1M context, Terminal-Bench 87.9 that sits rightwitter
  3. [3]JulianGoldieSEO — DeepSeek V4 Pro just landed 0.1 behind the top model on its own Terminal Bench numbers. And the price gap is wild. → 87.9 on Terminal Bench twitter
  4. [4]CyberMonk0x — DeepSeek V4 Pro正式版昨晚悄然上线 Agent能力暴涨，DeepSWE从12.8飙至62.7 多项测试逼近甚至追平Claude Fable 5 1.6T参数 + 1M上下文，API价仅3元/百万Token输入 开源旗舰性价比拉满，闭源高端终于有了真对手。 #Deeptwitter
  5. [5]kcp_kn — DeepSeek V4 ProそこまでFable 5とベンチマーク上は差がないのにこの安さ。 $0.435 M/Input $0.87 M/Output — quoting @AndrewCurran_: As yet unverified DeepSeek V4 Pro bentwitter
  6. [6]intern_11 — The new DeepSeek V4 Pro is even crazier than I expected. Launched August 13 at midnight. Model version: DeepSeek-V4-Pro-0813. 1 million toketwitter
8.11

Menlo Research launches $20,000 open-source humanoid robot DIY kit

  • Singapore-based Menlo Research (branded as Asimov Inc.) has launched the Asimov 1, an open-source reference humanoid robot measuring 120cm tall and weighing 35kg, with 25+2 degrees of freedom, priced at $20,000 as a complete DIY kit delivered to the buyer's door. [3][4][5]
  • The Asimov 1 is explicitly positioned as an educational platform for builders, engineers, and researchers; assembly typically requires over 100 hours and gives users full access to hardware and software so they can inspect every joint, sensor, and circuit. [1][2][5]
  • After assembly and training of the appropriate control policy, Asimov 1 can walk, balance, dance, and recover from stumbles; because hardware and software are fully open source, users can modify, repair, and customize the robot's capabilities. [5]
  • Menlo Research negotiates bulk pricing on core components so that purchasing the kit costs less than self-sourcing parts individually; the company notes final price may vary by destination country based on tariffs and shipping. [1][2][3]
  • Menlo Research also built a browser-based simulator called Digital Asimov that lets users explore the robot's design and test control ideas before working with physical hardware, at no cost and with no setup required. [5]
Menlo ResearchAsimov Inc.Asimov 1Digital AsimovSingaporeopen-source humanoid roboteducational robotics platformDIY robot kitrobot simulationhardware and software transparencycontrol policy traininghumanoid locomotionresearch robotics
Sources
  1. [1]chris_j_paxton — Asimov is looking really slick - amazing price for a full humanoid robot that's totally open source — quoting @asimovinc: Asimov 1 is an ope★twitter
  2. [2]DynamicWebPaige — 🙌 Yes, this! Open hardware, open software and models, and DIY kits are the only way to do robotics: — quoting @asimovinc: Asimov 1 is an opetwitter
  3. [3]asimovinc — Asimov 1 is an open-source reference humanoid robot. It ships as a complete DIY kit. $20,000 total, delivered to your door. https://t.co/OzLtwitter
  4. [4]asimovinc — Asimov 1 is an open-source reference humanoid robot. It ships as a complete DIY kit. $20,000 total, delivered to your door. https://t.co/OzLtwitter
  5. [5]Humanoidguide — Singapore’s Menlo Research built Asimov 1, a 120cm, 35kg open-source humanoid you assemble yourself for $20,000. It walks, balances, avoids twitter