Ai Model

Google DeepMind chief plans early release of Gemini 4 model: Google DeepMind's new chief, Koray Kavukcuoglu, plans to release Gemini 4 much earlier than the end of the year. The model is currently in post-training and is running internally within the Antigravity coding tool. Kavukcuoglu shifted focus toward trustworthy agents, calling AGI "not the right conversation."

AI Models • The Decoder

Black Forest Labs releases FLUX 3 Action robotics model with 7B parameters: Black Forest Labs has introduced FLUX 3 Action, an open-world-action robotics model that predicts robot actions from camera feeds. The 7-billion-parameter model set a record on the RoboLab-120 benchmark. It also runs up to 3.95 times faster than the previous top-performing model.

robotics • The Decoder

Anthropic Claude Opus 5.5 breaks GPT-6 benchmark during evaluation: Anthropic's Claude Opus 5.5 was benchmarked against OpenAI's GPT-6 Sol. During the evaluation, Opus 5.5 continuously built, judged, fixed, and iterated on tasks until the benchmark itself broke down.

LLMs & Model Evaluation • The Neuron

Anthropic went CRAZY (Opus 5.5): The video features a series of performance tests evaluating Anthropic's Claude 3.5 Opus artificial intelligence model. Guest Thariq joins to analyze the model's capabilities and benchmark results. The discussion highlights how the model compares to previous iterations and competitor offerings.

Technology • Matthew Berman

OpenAI launches GPT-6 Sol and Luna models with lower costs and fewer errors: OpenAI has released GPT-6 Sol and GPT-6 Luna less than three months after the launch of GPT-5.6. The company claims that GPT-6 Sol makes about half as many mistakes as GPT-5.6 Sol, while GPT-6 Luna matches the performance of GPT-5.6 Sol at approximately 1% of the cost.

Artificial Intelligence • Techmeme

Anthropic releases Opus 5.5 with lower pricing and Fable-level performance: Anthropic has launched its Opus 5.5 model, offering lower pricing and performance comparable to its Fable model. The company described the new release as the strongest-performing model it has tested to date.

Artificial Intelligence • TechCrunch AI

Li Auto releases embodied AI trilogy including ME-Brain-1.0: Li Auto's Foundation Model team has released its embodied AI trilogy, consisting of ME-Brain-1.0, the ME-U0 unified world-action model, and ME-VLM cognitive models. The ME-VLM release includes a 35B-A3B model and a 4B edge SKU designed for the M100 SoC. The release emphasizes system architecture over vehicle branding.

Artificial Intelligence • Pandaily

Inspur launches SD200 Ultra supernode with 128 domestic AI chips: Inspur Information has launched the MetaBrain SD200 Ultra supernode featuring 128 tightly coupled domestic AI chips and 8 TB of unified accelerator memory. The system reportedly achieves a latency of under 5.85 ms per token when running the 2.8-trillion-parameter Kimi K3 model.

Artificial Intelligence • Pandaily

NVIDIA details Confidential Computing to secure LLM inference workloads: NVIDIA has detailed its Confidential Computing technology designed to secure large language model inference. The system aims to protect sensitive information and proprietary model context during high-performance production AI workloads across personal, enterprise, and regulated domains.

NVIDIA • NVIDIA Generative AI

Open Jev Models Are Here!!: A review of seven open-source Jev-style AI models examines their capabilities and performance. The evaluation utilizes tools and repositories including JevBench, SemIf, Bespoke-Nimble-9B, Decider, and Alex Wortega's OpenJev.

Technology • Sam Witteveen

StepFun releases Step 5 Preview open-source model with 27B active parameters: StepFun has released Step 5 Preview, an open-source model featuring 27 billion active parameters. The model ranks among the top two open-source models in recent evaluations, demonstrating strong performance despite its compact active parameter size.

AI Models • 量子位

Researchers introduce two-step test-time training protocol for VARC models: Researchers have introduced a novel two-step test-time training (TTT) protocol to improve rule induction in Vision ARC (VARC) models. The method first finetunes only the task embedding representing the transformation rule, and then freezes it to finetune the backbone.

AI Research • arXiv

LogicTrack framework uses formal logic solvers to audit LLM reasoning steps: Researchers have proposed LogicTrack, a neuro-symbolic framework designed to verify the logical validity of intermediate reasoning steps in large language models. The system auto-formalizes each reasoning step into symbolic representations and verifies them using automated theorem provers.

AI Research • arXiv

DeepSeek CEO calls training on Huawei chips one of its biggest bets: DeepSeek CEO Liang Wenfeng told investors that training its models on Huawei chips is one of the company's biggest bets. Huawei is reportedly scheduled to deliver its next-generation training chips to DeepSeek in the fourth quarter of 2026 or the first quarter of 2027.

Artificial Intelligence • Techmeme

Unitree releases partial weights for 6-billion-parameter UnifoLM model: Unitree has released partial weights and data for UnifoLM-WLA-1.0, its 6-billion-parameter world-language-action model for humanoid robots. The model covers 64 tasks, though post-training code and full datasets are still planned for future release.

humanoid robots • Pandaily

RBS-Attention training-free method optimizes long-context LLM prefill: To address the prefill bottleneck in long-context large language model inference, researchers proposed RBS-Attention, a training-free sparse-prefill method. The approach uses a centroid base branch to capture average relevance and a rescue branch to identify blocks at risk of underestimation due to mean dilution.

AI Research • arXiv