Anthropic Hits $900B Valuation with Opus 4.8 Launch Amid CNN Lawsuit Against Perplexity
Model Releases & Updates
Anthropic Releases Opus 4.8: Anthropic has launched Opus 4.8, a new model that reportedly outperforms competitors like GPT-5.5 and Gemini 3.1 Pro across various benchmarks. The update shows significant improvements in reasoning, agentic coding, and handling complex tasks.
StepFun Introduces Step 3.7 Flash: StepFun has released Step 3.7 Flash, a 196B parameter multimodal Mixture of Experts (MoE) model designed for local deployment. It features a built-in Vision Transformer and outperforms other flash-tier models in benchmarks such as SWE-Bench Pro.
Liquid AI Launches Edge-Optimized LFM2.5-8B-A1B: Liquid AI’s new LFM2.5-8B-A1B is an advanced 8B parameter model designed for edge devices and laptops. It features a 128K context window, heavy reinforcement learning, and improved tokenization for non-Latin languages.
Industry Milestones & Partnerships
Anthropic Surpasses OpenAI in Valuation: Anthropic has officially become the world’s most valuable AI startup with a valuation of $900 billion, overtaking OpenAI. This milestone reflects massive investor confidence in Anthropic’s growth and its current positioning within the AI industry.
Airbus and Mistral AI Form Strategic Partnership: Airbus has partnered with Mistral AI to deploy sovereign AI technologies across its commercial, defense, and space sectors. The collaboration aims to enhance automation and design simulations using Mistral’s specialized AI solutions.
Legal & Service Issues
CNN Sues Perplexity AI Over Copyright Theft: CNN has filed a lawsuit against Perplexity AI, alleging that the company used its content without authorization. This case underscores the intensifying legal friction between major media organizations and AI-driven platforms regarding intellectual property.
Google Gemini Faces Criticism for Performance Degradation: Users are reporting a decline in the quality of Google’s Gemini AI, citing restrictive rate limits and increased hallucinations. Many have observed that the models perform better when accessed through third-party platforms like Perplexity than via Google's official app.
Technical Optimizations
Zai Upgrades Inference Architecture for GLM-5.1: Zai has replaced its standard network setup with a new architecture called ZCube to run GLM-5.1 inference more efficiently. The upgrade resulted in a 15% increase in throughput and a 40.6% drop in tail latency while reducing hardware costs.
llama.cpp Update Reduces VRAM Consumption: A recent pull request for llama.cpp introduces an f16 mask for Flash Attention (FA). This feature allows users to save VRAM, enhancing the efficiency of the software when running large language models locally.