Open-Weight Models

AIAI-generated summary TIER IV detailed five open-source AI models for autonomous driving spanning camera-only systems and configurations adding LiDAR, radar, and high-definition maps. The lineup includes NVIDIA Alpamayo 2 Edge, AutoE2E, METEOR, OnePlanner, and LatentDrive for Autoware 2.0.

AIAI-generated summary Genspark and Baseten Applied Research announce the post-training of an open-weight model starting from Qwen3.5-397B. The companies state that their new model scores 0.759 on internal evaluations, which outperforms Opus 5.5 and serves as the default production model for Genspark's Deep Research agent.

AIAI-generated summary Ollama is an open-source platform that enables users to run large language models locally on macOS, Linux, and Windows without relying on cloud services. The software supports models such as Llama 4, Qwen 3, Gemma 3, Phi-4, DeepSeek-R1, and Mistral, and allows customization through a Modelfile.

AIAI-generated summary NaiveAI has developed an open-weights model called Naive-N0.5-Flash for AI coding tasks, designed to generate a high number of tokens per second. The company states that the model aims to speed up engineering and research workflows to help teams iterate and solve complex problems faster.

AIAI-generated summary MiniMax announced the launch of MiniMax H3-Image, an open-weight model capable of 4K image generation in 10 seconds with a 16 million pixel resolution. MiniMax states the model features precise prompt following, editing capabilities, and support for up to 16 reference images.

AIAI-generated summary Unite.AI reports that private artificial intelligence requires infrastructure that can evolve with workloads, power, cooling, and compute needs. Linux Foundation research shows 63% of surveyed companies used an open artificial intelligence model.

AIAI-generated summary Ant Group embodied AI company Robbyant has open-sourced LingBot-Touch, a tactile foundation model designed to give robots advanced touch perception using a dataset of nearly 5,000 hours of real-world tactile data. Robbyant says the model supports magnetic tactile, visual tactile, and piezoresistive sensing, which could improve fine manipulation tasks such as turning pages, unscrewing bottle caps, and identifying Mahjong tiles.

AIAI-generated summary MiniMax announced the release of MiniMax H3-Image and its model weights to the open-source community. MiniMax says the new image model generates 4K by 4K images in 10 to 20 seconds and supports instruction following.

AIAI-generated summary Devon Thomas founded a company named Abliteration AI that removes safeguards from open-weight models and makes them available to anyone. Eric Hartford, an AI scientist based in Charlotte, North Carolina, developed an uncensored model called Dolphin by modifying Meta’s base Llama.

AIAI-generated summary Production inference requires engineering teams to balance model choice, deployment approach, and infrastructure capacity to meet specific latency and throughput targets. Managed compute in Microsoft Foundry utilizes hardware such as AMD Instinct MI300X GPUs to let organizations run open models without managing the entire serving stack.

AIAI-generated summary NVIDIA released Dynamo-optimized deployment recipes for open models including A.X K2, K-EXAONE 2.0, Solar Open 2, and Motif-3. NVIDIA says the recipes optimized for B200 GPUs deliver up to two times performance and help developers bring open models into production faster.

AIAI-generated summary The author used LM Studio and the free iPhone app Locally to access a local AI setup running Gemma 4 26B on a desktop PC. LM Studio uses a feature called LM Link, built on Tailscale, to connect mobile devices to models running on a personal computer.

AIAI-generated summary Nace published a technical report regarding Drex, which is a family of four open-weight decision models designed to return probabilities for permitted choices without generating word-by-word explanations. Nace reports that when Drex Max, a 27-billion-parameter model, states it is at least 93 percent sure, the company reports it is right 97 percent of the time.

AIAI-generated summary Vals AI reports that autonomous AI agents can infiltrate realistic corporate networks without human assistance. The Vals CyberOffense benchmark shows that open-weight models achieve competitive cyber performance compared to closed systems while costing significantly less.

AIAI-generated summary Nous Research fine-tuned Alibaba's Qwen3.8-27B on 53 million tokens to create Arete-27B and reduce Chinese state propaganda rates from 22 percent to under one percent. According to Nous, the resulting open-weight model maintains its general capabilities while dropping the propaganda rate and increasing factual accuracy.

AIAI-generated summary Amazon Web Services, Inc. announces that Kimi K3.1 Pro from Moonshot AI is now available on Amazon Bedrock. Moonshot AI states that Kimi K3.1 Pro is a natively multimodal agentic model featuring a 1-million-token context window.

AIAI-generated summary Research from Vals shows that open-weight artificial intelligence models are nearly as capable of hacking computer networks as leading closed models. According to Vals, OpenAI's Daybreak completed 81 percent of tasks while the Chinese open-weight model DeepSeek completed 73 percent.

AIAI-generated summary Nace AI has released a report detailing the training process for Drex, which is a family of open-weight AI models designed to select the most likely answer from a fixed list. The model called Drex v1.5 was trained using verified answers for tasks including contract reading and phishing detection, and it can be tested within Nace Console.

AIAI-generated summary Thinking Machines Lab released Inkling 2 Small, a model with 594 billion parameters that will have its weights released next week. Artificial Analysis stated that the model scored a 31 on the Artificial Analysis Intelligence Index and noted its pricing is somewhat expensive.

AIAI-generated summary Google introduced Gemini 4 Carbon and Google DeepMind previewed Genie 4, while Mistral published the weights for its Large 4 model. Anthropic filed its IPO paperwork publicly and OpenAI secured about $11 billion in funding for its latest round.