Nvidia opens Flora to cut robot-video labeling costs
Nvidia has open-sourced Flora, a robotics data engine that labels 200,000 hours daily and claims a 125-fold cost advantage over GPT-6 Astra.
Open Source — Category
Moonshot launches Kimi K3.1 Pro with 3T-class scale
Moonshot’s new Kimi K3.1 Pro combines a 1M-token context window, tool use and multimodal workflows with immediate API access.
StartLux Open-Sources Prism for Faster Decision Models
StartLux releases Prism, an open inference engine for its calibrated decision models, with reported speedups of up to 5.3x in selected workloads.
IBM Releases Open Granite Time-Series Ensemble Recipe
IBM Research publishes an open forecasting recipe that combines four Granite time-series models for zero-shot and probabilistic prediction.
Recursive Releases Flint, a 40B Open Coding Model
Recursive has released Flint under Apache 2.0, pairing a sparse 40B architecture with local deployment for coding agents.
Tencent’s WorkForge Scales Verifiable Training for Work Agents
Tencent researchers introduce WorkForge, a framework that turns real-world files and workflows into verifiable environments for long-horizon agents.
DeepSeek Open-Sources the Inference Stack Behind V4.1 Flash
DeepSeek released DeepInfer and deepseek-recipe, opening the serving software behind its latest model and lowering barriers to large-scale deployment.
Tenstorrent raises $1.5 billion for open-model chips
Tenstorrent reached a $6.2 billion valuation after raising $1.5 billion to scale inference hardware for open-weight models and video generation.
Researchers show LLM agents can hide DNS exfiltration
A new arXiv paper describes a model-weight backdoor that can covertly activate DNS data theft, exposing a supply-chain risk for agent deployments.
DeepSeek Releases V4.1 Flash-1017 With Open Weights
DeepSeek has released a refreshed open-weight model and opened its search and fetch APIs, extending its push into low-cost agent infrastructure.
Reflection Releases 501B Beam Model Under Apache 2.0
Reflection has released Beam’s weights and tooling, opening a 501-billion-parameter coding model built for long-horizon agent workloads.
Meta releases 816B-parameter Muse Spark Open
Meta has released its largest open-weight model, pairing a 1M-token context window with coding, multimodal and long-running agent capabilities.
Nvidia Opens 2.8T-Token Pipeline for Indic AI Models
Nvidia details a 2.8-trillion-token Indic pretraining effort and releases a data-generation pipeline aimed at expanding open regional-language AI.
Perplexity open-sources SPACE for safer AI agents
Perplexity has open-sourced SPACE, a sandbox platform that isolates computer-using agents while preserving task state through snapshots.
Skywork Releases Open Matrix-Game 4.0 for Real-Time Worlds
Skywork says Matrix-Game 4.0 is an open-source video model game that runs in real time on one consumer GPU.
StepFun Opens Step 5 Preview Weights for Long-Horizon Agents
StepFun has released Step 5 Preview as open weights, pairing a 600B sparse MoE model with a new harness for sustained agent work.
Alibaba introduces elastic RL for long-horizon agents
Alibaba’s QwenGyre framework reallocates training resources across difficult rollouts, reporting gains on long-horizon software-engineering tasks.
Nvidia Weighs Deeper Tie-Up With Reflection AI
Nvidia is reportedly considering an acquisition, acqui-hire or larger investment in Reflection AI, tightening its control over open-model talent.
vLLM Opens Multimodal Decision Models for AI Routing
vLLM Semantic Router released Decision 3.0 models from 0.6B to 26.09B parameters that return structured choices instead of generated prose.
Alibaba Releases Open-Weight Qwen-Image-2.1-Turbo
Alibaba’s Qwen team released an open-weight image model designed to generate and edit visuals in just eight denoising steps.
Anthropic launches Cyber Mission and free OSS Scanner
Anthropic is pairing frontier models with critical-infrastructure support and free vulnerability scans, turning AI cyber defense into a public deployment program.
Liquid AI Releases Open d1 Models for Edge Decisions
Liquid AI has released open-weight multimodal decision models designed to make fast structured choices on phones, PCs and edge hardware.
Perplexity Releases Multimodal Late-Interaction Embeddings
Perplexity’s new 0.6B and 9B embedding models match text and images in one space, targeting more precise retrieval across rendered documents.
Google Releases Multimodal EmbeddingGemma 2
Google released an open 740-million-parameter embedding model that maps text, code, images, audio and video into one space for local retrieval.
Meta Open-Sources Rebalancer for Large-Scale Assignment
Meta has released the C++ and Python solver used across dozens of large-scale allocation problems involving servers, traffic, and machine-learning workloads.
Mistral Opens Preview of One-Trillion-Parameter Large 4
Mistral opened a public preview of its one-trillion-parameter multimodal model, pairing frontier claims with a promise of open weights later this month.
Reflection opens 501B Beam model for coding and agents
Reflection AI has released Beam, a 501-billion-parameter open-weight model that targets coding, reasoning and long-horizon agent workloads.
Aleph Alpha Releases 78B Kolibri Open-Weight Model
Aleph Alpha releases a German-English MoE model with 78.1 billion parameters, 3.46 billion active per token and Apache 2.0 licensing.
BootLoops Open-Sources Tools for Exact Science Agents
BootLoops releases an open toolkit that gives AI agents certified computational tools for high-precision physics and quantitative research.
Meta Opens Muse Gadget Code for DIY AI Hardware
Meta is opening Muse gadget software for ESP32 boards and Raspberry Pi devices, extending its personal agent beyond phones and computers.
Prime Intellect Launches Serving Layer for Open Models
Prime Inference offers serverless and reserved hosting for frontier open models, linking deployment traffic with the company’s continual-training loop.
Cloudflare Releases Open-Weight Clef Decision Models
Cloudflare released two open-weight models that return calibrated decisions instead of text, targeting faster and cheaper control loops for AI agents.
DeepSeek Opens Harness Desktop Agent for Local Work
DeepSeek has released Harness for macOS and Windows, combining local file access, coding, background tasks and extensible agent plugins.
AWS Releases Open Strands Decider 2B for Agent Decisions
AWS has open-sourced a 2-billion-parameter decision model that returns calibrated choices in roughly 115 milliseconds for agent workflows.
DeepSeek Open-Sources Ascend Kernels for V4.1 Inference
DeepSeek has released optimized attention kernels for Huawei’s Ascend 950, showing how open software can improve China’s alternative AI stack.
NVIDIA Unveils Hardware-Enforced Safety for AI Agents
NVIDIA introduced OpenShell and Sentry, a software-and-hardware stack designed to constrain, monitor and rapidly quarantine autonomous agents.
Holo4 Targets Generalist Computer-Use Agents
H Company’s Holo4 pushes computer-use agents toward broader interface skills, extending the open model race beyond chat and coding.
Black Forest Labs Releases Open-Weight FLUX 3 Action
The 7B model predicts video and robot actions together, placing a visual-generation lab inside the race to build open physical-AI systems.
StepFun open-sources Step Code for end-to-end development
StepFun released Step Code, an MIT-licensed coding agent that edits, tests and ships software from one CLI with strong long-horizon results.
Alibaba Plans Qwen Model Reaching 10 Trillion Parameters
Alibaba says future Qwen models could reach 5–10 trillion parameters as it links model scaling to domestic chips, cloud capacity and agent deployment.
AWS Releases Open-Source Strands Agent Harness
AWS launches Strands Harness as a ready-made open-source agent stack, claiming lower token use at comparable task accuracy.
NVIDIA Open-Sources SoL-Pi for Leaner Coding Agents
NVIDIA’s SoL-Pi extension uses automated research loops to reduce coding-agent token traffic by up to 49% while preserving task performance.
Xiaomi Open-Sources MiMo-V2.6 After Large-Scale RL
Xiaomi releases MiMo-V2.6 Pro and Flash with open weights, RL environments and training tools for agent research.
China Telecom Opens Xing4.0 Agent Model Weights
China Telecom released Xing4.0-29B-A4B, an Apache-licensed sparse model trained on Ascend hardware and built for long-context agent workflows.
Anthropic Open-Sources Claude’s Biology Model Optimizations
Anthropic says Claude optimized more than 30 open biology models, delivering roughly fourfold speedups and a lower-memory operating mode.
StepFun releases five-model StepAudio 3 family
China’s StepFun releases five StepAudio 3 models spanning realtime speech, recognition, synthesis, audio generation, and music through one platform.
ZGCM-1 Opens a Full 7B Training Pipeline
Seven doctoral students released a 7B model with weights, data recipes, checkpoints and logs, offering an unusually complete blueprint for agentic model training.
Shanghai AI Lab Releases Intern-S2-Preview-397B for Science
Shanghai AI Lab introduces Intern-S2-Preview-397B, a 397B multimodal model built for scientific reasoning, long-horizon agents, and research workflows.
NVIDIA Open-Sources OSMO for Physical AI Workflows
NVIDIA releases OSMO, a YAML-driven orchestrator linking robot data, simulation, training, and hardware testing across mixed compute.
Sakana Fugu Max Targets Frontier AI at Lower Cost
Sakana AI released Fugu Max and Fugu Ultra v2, using model orchestration to pursue cheaper agent performance and higher autonomous-task scores.
DeepSeek Releases V4.1 Flash With Million-Token Context
DeepSeek has launched V4.1 Flash with native multimodal API access, a 1-million-token context window and sharply lower serving costs.
Tencent Hunyuan Open-Sources AuK for Speech and Audio Editing
Tencent Hunyuan and research partners have released AuK, an open-source foundation model for controllable speech and audio generation and editing.
Mistral Raises €3 Billion to Expand Sovereign AI
Mistral’s record European funding round gives its open-weight strategy unprecedented capital, while tying Samsung and Europe more closely to frontier AI.
Nvidia Agrees to Acquire Hugging Face for $12.9B
The chipmaker is buying the leading open-model hub while promising continued support for rival hardware, clouds and models.
Perplexity Open-Sources Lily for Faster Qwen Inference on Macs
Perplexity has released the Rust and Metal source, tests and reproduction materials for an engine it says runs Qwen3.6-35B-A3B up to 1.35 times faster than MLX-LM on an M5 Max.
Google TimesFM-3 Adds Zero-Shot Multivariate Forecasts
The 330-million-parameter model forecasts related series and external variables without task-specific fine-tuning.
Microsoft Shrinks Pathology Models for Cohort-Scale AI
Open-weight GigaPath-Flash and GigaTIME-Flash sharply reduce compute and memory needs while retaining research performance.
Ollama Replaces Usage Windows With Per-Token Pricing
Ollama has replaced opaque GPU-time limits with published token rates and monthly credits across its paid cloud plans.
Together AI, HUMAIN Plan 250MW Saudi AI Data Center
The partners say the project will nearly triple Together AI’s capacity and support more than $5 billion in annualized revenue.
OpenClaw 2.0 Turns Personal Agents Into a Shared Platform
The open-source agent adds remote sessions, interactive dashboards and scoped secret handling in a major architectural overhaul.
Z.ai Opens GLM-5.3 Weights With Stronger Cyber Skills
The 743-billion-parameter coding model is now downloadable, turning its reported cyber gains into a broad access question.
Tencent Opens Hy4 Preview With One-Million Context
Tencent has released Hy4 Preview’s open weights, pairing a 770-billion-parameter MoE design with a one-million-token context window.
Z.ai Releases GLM-5.3-Flash With Open Weights
The 320B mixture-of-experts model pairs native multimodality and one-million-token context with aggressively priced hosted inference.
Nvidia Reportedly Nears $12.9B Hugging Face Deal
Conflicting reports say Nvidia has agreed, or is close, to acquire the central hub of the open-model ecosystem for $12.9 billion.
Qwen Opens Flash-Next as a Preview of Qwen4
Alibaba’s 125B-parameter multimodal model activates 6B parameters per token and introduces the architecture intended for Qwen4.
MiniCPM Passes 50 Million Downloads in Edge AI Push
ModelBest says its open MiniCPM family has surpassed 50 million downloads as the Chinese lab prioritizes efficient, on-device AI.
Gradio Adds Visual Workflows That Deploy as APIs
Gradio’s new Workflow interface turns typed AI pipelines into inspectable canvases, callable REST endpoints and deployable applications.
Hugging Face Explores Sale at $13 Billion Valuation
The open-model hub is reportedly testing buyer interest, raising questions over who could own a critical layer of AI infrastructure.
Microsoft’s Skala 1.1 Broadens Predictive Chemistry
Microsoft’s updated neural chemistry functional improves accuracy and moves into software already used for molecular simulation.
Alibaba’s Qwen Models Cross Three Billion Downloads
Qwen’s download milestone signals that the center of gravity in open-weight AI is shifting toward China’s developer ecosystem.
Cohere Opens 2.4B Vision Model for Edge Deployment
North Micro Vision packages document and visual understanding into a 2.4B model with downloadable weights and an Apache 2.0 license.
Alibaba Opens 2.4-Trillion-Parameter Qwen3.8 Weights
Alibaba has released its largest Qwen model’s weights, giving independent operators a frontier-scale foundation for coding and agents.
Ant Group Releases Compact Ling 3.0 Tiny Model
The 7.9B-parameter mixture-of-experts model targets economical local and agent deployment, activating only 1.3B parameters per token.
LTX-2.5 Brings Adaptive Compute to Open Video Models
LTX’s new open-weight world model adds adaptive rendering, native 4K HDR output and stronger multi-shot control for local video production.
Mistral Opens Its AI Platform to Third-Party Models
Mistral will host rival open models, beginning with Z.ai’s GLM-5.2, while adding regional inference and reserved European capacity.
Nvidia Releases Nemotron 3.5 Lightning for Agents
The open 30B mixture-of-experts model activates 3B parameters per token and ships with training data, recipes and a new model router.
Meta's Muse Glimmer debuts mid-pack in Arena boards
Arena's first human-preference scores put Meta's new Apache-2.0 model at #77 in Code Arena WebDev and #97 in Text Arena, 26th and 24th among open models.
Zhipu's ZCode hits 1M users, resets all usage limits
Z.ai says its agentic coding environment passed one million users and wiped usage counters for every GLM Coding Plan subscriber as a thank-you.
Zuckerberg manifesto resets Meta on open weights
In a 6,500-word essay Zuckerberg recommitted Meta to open models, attacked closed rivals for concentrating power, and floated auctioning superintelligence compute.
Meta open-sources 30B Muse Glimmer under Apache 2.0
Meta Superintelligence Labs released its first open model — a 30B agentic multimodal system under Apache 2.0 that runs on a single consumer GPU.
Ling 3.0 Flash sets open-weights efficiency frontier
Artificial Analysis scored Ant Group's 124B open-weights model at 38, placing its 5B active parameters ahead of every flash-tier rival.
OpenAI, Vercel and rivals ship Agent Plugins 1.0 standard
A vendor-neutral format bundles Agent Skills and MCP server configs into one portable plugin, supported at launch by ChatGPT, Codex, Cursor, Copilot, Kiro and VS Code — with Google joining the steering committee the same day.
DeepMind open-sources WeatherNext cyclone forecasting model
A Nature paper puts Google DeepMind's WeatherNext Cyclones ahead of operational systems by over a day of lead time, with code and weights released on GitHub.
Huawei's openJiuwen ships agent swarm into bank production
The open-source agent platform released an enterprise-grade distributed swarm architecture, first deployed in production at China's Postal Savings Bank.
Cursor open-sources its MoE training megakernel MoK
Mixture-of-Kittens fuses all MoE communication and compute into one deterministic Blackwell kernel, lifting Cursor's training throughput 1.41x.
JD open-sources 16B real-time video editing model
JD.com released Apache-2.0 weights for JoyAI-Video-Edit, which edits 720p video at about 30 frames per second on a single NVIDIA B200.
Mistral open-sources Shieldstral, a 3B moderation model
The Apache-2.0 guard model takes safety policies as plain-language questions at inference and runs on a single 16GB GPU.
Sand.ai open-sources MAGI-2, a 114B MoE video model
Beijing lab Sand.ai released Apache-2.0 weights for a 114-billion-parameter sparse video model that generates 1080p clips with synchronized audio.
MiniMax H3 tops Video Arena as No.1 open video model
Video Arena placed MiniMax-H3 first among open models in both text-to-video and image-to-video, some 280 Elo points clear of the next open weights entry. The weights are now public, but the licence restricts open-weight use in the US, EU, UK and South Korea.
Genspark open-sources GenOffice suite for PC and Mac
Genspark released GenOffice, which it calls the first full-featured open-source AI office suite, free and ad-free on Windows and macOS under Apache-2.0.
MiniMax releases open weights for its H3 video model
MiniMax published downloadable weights for H3, its omni-modal 2K video model, making a top-ranked video generator locally runnable for the first time.
Sakana AI launches Namazu, a Japanese-tuned LLM API
Tokyo lab Sakana AI opened commercial access to Namazu, a Japanese-specialised model built by fine-tuning Moonshot's open-weight Kimi K2.6.
Supabase open-sources an eval suite for coding agents
Supabase released Evals, an Apache-2.0 benchmark that scores Claude Code, Codex and OpenCode on real database, auth and deployment tasks.
JetBrains open-sources runtime LLM macros for Kotlin
JetBrains Research released KotlinLLM under Apache 2.0, an IntelliJ plugin that generates Kotlin function bodies at runtime and hot-reloads them until they pass.
Thinking Machines open-weights 276B Inkling-Small
Two weeks after its 975B flagship, Thinking Machines released a 276B Apache-2.0 sibling with 12B active parameters and full text, image and audio reasoning.
Together AI open-sources ThunderAgent agentic inference layer
ThunderAgent schedules whole agent workflows instead of single calls, reporting 2x throughput and 6x lower latency than SGLang on one 8xH100 node.
Liquid AI open-weights two CPU-fast 8K text encoders
LFM2.5-Encoder-230M and 350M bring an 8,192-token window to BERT-class models and cut CPU latency roughly threefold at full length.
OpenAI open-sources its Codex Security scanning CLI
The Apache-2.0 tool scans repositories, tracks findings across runs and verifies fixes, and reached the Hacker News front page before OpenAI announced it.
Anthropic says it never sought a ban on open-weights models
Responding to months of speculation, Anthropic published its open-weights position: no bans, but mandatory pre-release safety testing for any sufficiently capable model.
China's domestic chips run 2.8T Kimi K3 on day zero
Alibaba Cloud says its Zhenwu M890 supernode is the first Chinese system to run Moonshot's 2.8-trillion-parameter model; Huawei Ascend also claims day-0 support.
Moonshot's 2.8T-parameter Kimi K3 weights land on Hugging Face
Moonshot AI has published Kimi K3's full 2.8-trillion-parameter weights under the Kimi K3 License — the largest open-weight model release to date.
Moonshot open-sources AgentENV, the sandbox fleet behind K3
The Kimi team and kvcache-ai released the Firecracker-based environment system used for Kimi K3's agentic RL training under an MIT licence.
Moonshot opens PerceptionBench; no model clears 60%
The Kimi team released a benchmark that strips reasoning out of multimodal evaluation and grades whether a model actually saw the image correctly.
Nvidia-led Open Secure AI Alliance launches with 50+ members
Nvidia's Open Secure AI Alliance lists Microsoft, IBM, Hugging Face and others among 50-plus inaugural partners; separately, the open-weights policy letter has grown to 50 signatories including Google and OpenAI, with Anthropic and Amazon still out.
Kuaishou's KAT-Coder-V2.5 tops PinchBench with open Dev variant
Kuaishou's KwaiKAT team says its new agentic coder, trained by RL in 100,000+ verifiable repository environments, edges Claude Opus 4.8 on PinchBench.
Open Dreamer rebuilds Dreamer 4 in JAX with full training recipe
Research group Reactor has released a JAX/Flax reproduction of DeepMind's Dreamer 4 pipeline, publishing the full recipe and the stability fixes that made it train.
US said to favor selective bans on Chinese open models
Reporting indicates the White House is leaning toward targeting specific Chinese open-weight models on security grounds rather than a blanket prohibition.
Open-weights letter grows as Google and OpenAI sign on
The July 24 open-weights letter has expanded beyond Nvidia, Microsoft and Meta, with updated signatories now including Google and OpenAI while Anthropic remains absent.
OpenAI and Google join expanded open-weights letter
OpenAI and Google have joined the expanding open-weights appeal to Washington, making Anthropic the most visible major frontier-lab holdout.
StepFun brings Attention-FFN disaggregation to vLLM
StepFun says it is open-sourcing its Attention-FFN disaggregation work with the vLLM team, Ant Group and FastAFD, pushing a key MoE-serving efficiency technique into the community.
Andrew Ng releases OpenWorker, an open-source desktop agent
Andrew Ng shipped an MIT-licensed, local-first desktop agent that plans tasks across files and apps and hands back finished deliverables instead of chat.
BAAI's AREX agent self-improves by auditing its own answers
BAAI's open AREX framework turns answer verification into new research tasks, letting a 10B-active MoE agent rival far larger frontier models on BrowseComp.
DeepSeek V4 goes GA with 1M context as legacy APIs retire
DeepSeek's flagship V4 line exits preview: legacy chat and reasoner endpoints shut down today as new peak and off-peak API pricing takes effect.
Cisco open-sources tiny Antares models to locate code flaws
Cisco's Foundation AI arm releases 350M and 1B open-weight models that trace known vulnerabilities to specific files, nearing frontier accuracy at a fraction of the cost.
Poolside's Laguna S 2.1 tops open-weight agentic coding class
Poolside released Laguna S 2.1, a 118B MoE coding model with 1M context under an open license, hitting 78.5% on SWE-Bench Multilingual with only 8B active params.
NVIDIA ships Cosmos 3 Edge, a 4B world model for on-device robots
NVIDIA open-sourced Cosmos 3 Edge, a 4B-parameter world model that reasons and generates robot actions locally, hitting 15 Hz real-time control on Jetson Thor.
Moonshot pauses Kimi K3 signups as demand outstrips GPUs
Moonshot AI halted new Kimi K3 subscriptions after demand surged sixfold, as third-party tests put the open-weight model within reach of Claude Fable 5.