THE COLLECTION
EVERY SIGNAL, SORTED.
Search the whole shelf by topic, tool, term, or plain old curiosity.
France, Mistral AI, Mistral Large 4, Le Chonk, Open Weights, Coding Agents, Cybersecurity, Self-Hosting, Model Licensing, Digital Sovereignty, Sandboxing and Human Oversight
Mistral released Le Chonk's weights. Put the cyber model in a sealed room
Mistral Large 4 is no longer just a hosted preview and a promise. The one-trillion-parameter model can now be downloaded and run under an organization's control. That freedom should begin with a locked sandbox, an artifact receipt and a very boring shutdown drill.
France and European AI, Mistral Large 4, Open Weights, Model Release Planning, Self-Hosting, Digital Sovereignty, Licensing, Infrastructure, Cybersecurity and Procurement
Mistral pinned Large 4 weights to October 31. Treat the files as a deployment gate
Mistral's trillion-parameter model works through an API today, but the self-hostable artifact remains an upcoming release. The calendar is useful. The files, license and reproducible deployment are the real event.
China, MiniMax, M3.1 Flash, Open Weights, Mixture of Experts, Multimodal Models, Local Inference, Quantization, Hardware, Licensing and Reproducibility
MiniMax opened M3.1 Flash weights. Portability needs a hardware receipt
MiniMax has published the weights for its 428-billion-parameter multimodal model. A 252 GB repository is downloadable, but portability begins only when someone records the exact artifact, license, hardware, runtime, memory, speed, quality and cost of a working deployment.
China, Open-Weight Models, Coding Agents, Multimodal AI, Model Evaluation, Inference Operations, Tool Use and Deployment Governance
DeepSeek made the model faster. The receipt still starts with the harness
DeepSeek says its October 17 update improves reasoning, coding, agents and vision while finishing work with fewer tokens. The useful question is not whether one score went up. It is whether the exact model, encoder, tools and hardware make your work cheaper without making it stranger.
United States, Open-Weight Models, Coding Agents, Model Deployment, Infrastructure, Evaluation, Safety, Licensing and Operational Control
Beam opened the loading bay. Running it is still heavy work
Reflection has released the weights, license, model card, safety report and deployment stack it promised. That makes Beam genuinely inspectable. It does not make a 501-billion-parameter agent model light, cheap or self-governing.
Liquid AI, Local Agents, Open Weights, Tool Permissions, Reward Hacking and Device Privacy
Liquid AI put a 24B agent on a laptop. The local claim needs a threat model
LFM2.5-24B-A2B can run an agent loop on consumer hardware with no cloud model API. That can improve privacy, latency and cost control. It does not make files, secrets, tools or external actions safe by default.
European AI, Open Weights, Multimodal Models, Digital Sovereignty, Model Evaluation, AI Infrastructure and Cybersecurity
Mistral put Europe's sovereignty claim behind an API for now
Mistral Large 4 is a one-trillion-parameter European model served from company-operated infrastructure. The preview is live, the weights are promised later, and the evidence needed to test digital sovereignty is still arriving.
Germany, Europe, Open Weights, Bilingual AI, Sovereign AI, Benchmarking and Local Deployment
Aleph Alpha opened Kolibri's weights, but kept the benchmark home field
Germany now has a downloadable bilingual reasoning model with a permissive weight license and unusually detailed training notes. The next test is whether outside teams can reproduce the performance claims on tasks the model maker did not design.
Decision Models, Edge AI, Open Weights, Reinforcement Learning and Cloud Platforms
Cloudflare made a fast decision model, then tied the training loop to its cloud
Clef and Clef-flash answer bounded questions with probabilities instead of prose. Their Apache 2.0 weights can leave Cloudflare, but the company's new improvement loop is designed to capture outcomes, fine-tune the model and redeploy it inside Workers AI.
Open-Weight Models, Sparse Attention, Coding Agents and Reproducible Evaluation
NaiveAI opened a 309-billion-parameter model, but the benchmark still belongs to its maker
Naive-N0.5-Flash exposes weights, inference code and a sparse route through a very large model. That is a useful release. The speed and capability scoreboard is still company-run, hardware-hungry and waiting for independent reproduction.
Belgium, Open-Weight Models, Local Cybersecurity and Software Assurance
Belgium's Aikido wants the security model to stay beside the code
Altar turns a 1.51-terabyte model into a 328-gigabyte local tool for guarded source code. The clever compression is real. So are the four H200 GPUs, custom license and unanswered production questions.