AI News Digest | OpenAI Codex Evolves into System-Level Operator; Claude Opus 4.7 Explodes; Ternary 1.58-bit Model Released
1. Core Model Releases & Technical Iterations
Claude Opus 4.7 Officially Launches with Performance Controversy Anthropic has released the new generation Opus 4.7, which performs excellently on Vending-Bench 2. The official subscription limits have been raised to compensate for the consumption of its “long thinking.” However, experts like Ethan Mollick point out that its “adaptive thinking” has flaws that cannot be manually intervened, often misjudging non-technical tasks as “low intelligence” requirements. (Source: https://x.com/andonlabs/status/2044808572299878565) (Source: https://x.com/emollick/status/2044864822076969268) (Source: https://x.com/ClaudeDevs/status/2044868953206612154)
Google Gemma 4 26B Now Available Gemma 4 26B officially lands on Vertex AI MaaS, providing serverless access. Meanwhile, the SGLang team demonstrated inference acceleration solutions for this model at Gemma Party. (Source: https://x.com/GoogleCloudTech/status/2044883532946825333) (Source: https://x.com/lmsysorg/status/2044888006730072171)
Ultra-Low Bit Model Bonsai Released The Bonsai series models (8B/4B/1.7B) based on ternary weights (1.58 bits) have been released. The 8B version is only 2GB in size and can run at 60 tps in the browser on M4 Max, marking a major breakthrough in local edge AI. (Source: https://x.com/xenovacom/status/2044834473874874519) (Source: https://x.com/mervenoyann/status/2044841709075411047)
2. AI Programming & Developer Tools
OpenAI Codex Evolves into “AI Operator” OpenAI has updated Codex, transforming it from a programming assistant to a “workspace system.” The new version supports parallel control of native Mac applications, image processing, and long-term automation tasks. Sam Altman called it “more useful than expected.” (Source: https://x.com/OpenAIDevs/status/2044828214867202519) (Source: https://x.com/sama/status/2044858862042591378)
Cursor Research Report: Developers Are Transitioning to “Managers” Cursor’s research found that the proportion of high-difficulty tasks has increased by 68% over the past year, with developers spending significantly more time on architecture design, code review, and documentation. (Source: https://x.com/cursor_ai/status/2044841478913130930)
LangChain Launches Reusable Evaluation Templates LangSmith now supports over 30 preset evaluation templates, covering “model as judge” and rule-based code evaluation. (Source: https://x.com/LangChain/status/2044824090205360393)
3. Multimodal Interaction & Specialized Applications
NVIDIA Lyra 2.0 Released This model can transform a single image into an explorable 3D world, supporting the export of 3D Gaussian, meshes, and physics engine parameters. (Source: https://x.com/minchoi/status/2044813729938694187)
Gemini Voice Interaction & Chrome AI Enhancement Gemini 3.1 Flash adds text-to-speech functionality, supporting tag-triggered tones; Chrome launches AI parallel search experience in the US, supporting sidebar comparison of web pages. (Source: https://x.com/GoogleAIStudio/status/2044852335848133113) (Source: https://x.com/NewsFromGoogle/status/2044836766838927525)
NotebookLM Personalization Upgrade Supports custom covers and creator notes. (Source: https://x.com/NotebookLM/status/2044896238341128272)
4. Robotics & Physical AI
Tesla FSD v14.3.1 Performance User testing achieved 2 hours of long-distance fully autonomous driving in “Mad Max” mode. (Source: https://x.com/Teslaconomics/status/2044830244243488922)
LeRobot & NVIDIA Robotics Day Hugging Face co-founder praised the open-source robot framework LeRobot; NVIDIA demonstrated the application of the Blackwell system in large-scale robot training at the University of Texas. (Source: https://x.com/ClementDelangue/status/2044868862865178934) (Source: https://x.com/NVIDIARobotics/status/2044832621969211666)
5. Industry Insights, Social Impact & Financial Updates
Anthropic Rumored to Reach Massive Agreement with AMD Rumors suggest that Anthropic will purchase AMD MI450 accelerators to diversify its computing power supply chain. (Source: https://x.com/HyperTechInvest/status/2044905842802082225)
X API Price Adjustment Notice Starting from 4/20, X API will increase the per-post price (to $0.015/post) and remove like and follow interface permissions for some low-tier plans. (Source: https://x.com/XDevelopers/status/2044919375153483812)
Funding Updates Eigen received a $15 million investment from Benchmark. (Source: https://x.com/paulscherer/status/2044812094953164914)
⚡️ Editor’s Note: This issue focuses on the battle of giants: OpenAI upgrades Codex to a system-level AI operator, countering Anthropic’s release of Claude Opus 4.7. Meanwhile, the release of Google Gemma 4 and the 1.58-bit ternary model Bonsai marks the rapid evolution of large models toward high energy efficiency and edge deployment.
Don’t want to fall behind in the information flood? Follow “Mao Mao Star” and click the star to pin it at the top. We monitor global AI developments around the clock for you, delivering only high-quality insights. Join AI geeks in witnessing the singularity of the intelligent era.
Be the first to know when I post cool stuff
Subscribe to get my latest posts by email.
Thanks for signing up! Check your email to confirm your subscription.
Whoops, we weren't able to process your signup.