AI News

AI changes every day,
catch up on it all on one page

The latest updates from AI companies, research institutions, and media around the world, automatically selected for you and updated around the clock.

Updated 11/17 05:25 (UTC+8) · refreshed hourly

10/23 · Friday5 items

OpenAI: Website updatesProductsAI Score67

OpenAI officially releases the Decisions API for low-latency, low-cost small decisions.

OpenAI announced that the Decisions API based on GPT-6 Luna is now generally available for bounded-answer problems such as classification, routing, and scoring. It is up to 10x faster than using GPT-6 Luna through the Responses API.

Why it's recommended:The official announcement states that the Decisions API is now generally available and provides speed comparisons with GPT-6 Luna, pricing, and GPT-Live integration, making it easier to evaluate its benefits for routing and classification scenarios.

Also reported by 3 other sources

10/22 · Thursday5 items

MiniMax:BlogProductsAI Score67

MiniMax Code desktop introduces Mini App functionality for long-running applications

MiniMax introduces Mini Apps in local sessions of the MiniMax Code desktop application (macOS / Windows). These are long-running applications with frontends and backends, installed as plugins and capable of two-way interaction with the Agent, allowing the Agent's work to remain on a persistent page rather than being scattered across conversations.

Why it's recommended:The official documentation explains the architecture, security boundaries, and community ecosystem of Mini Apps in detail, enabling developers to assess how to build long-running applications on MiniMax Code.

Kling AI@Kling_aiProductsAI Score67

Kling 4.0 launched on OpenArt, supporting first-, middle-, and end-frame guidance with up to 10 reference images

OpenArt announces the launch of Kling 4.0, offering video guidance using first, middle, and last frames, up to 10 reference images, text- or frame-based generation, and video editing. All-in-One Reference handles generation and editing with a single model. Pro and Wonder plans receive a 50% discount before November 20.

Why it's recommended:The cited content provides information on Kling 4.0's frame control, reference-image limits, and discounts on OpenArt, allowing readers to evaluate video-generation workflows.

Kling AI@Kling_aiProductsAI Score66

Kling 4.0 is now available on OnSolo.

Kling announced that Kling 4.0 is now available on OnSolo. The model supports generating 30-second native videos in a single pass, maintaining stability during fast motion and large camera movements, with stereo sound and more accurate lip synchronization. It also offers a 21:9 widescreen format, and a single task can combine up to 15 reference images and 10 keyframes to maintain consistency in characters, shots, and storylines.

Why it's recommended:The article details the specific capability changes in Kling 4.0 on OnSolo, helping creators assess its suitability for their video production workflows.

Claude:BlogProductsAI Score73

Anthropic launches the Claude Duet research preview, with Claude taking over daily work as a proactive agent.

Anthropic launched the Claude Duet research preview, an independent application that enables Claude to work as a proactive agent. It is available on iOS, Android, and claude.ai, with a waitlist open to Pro and Max users in the United States.

Why it's recommended:The original article details Claude Duet's operating mechanism, safety design, and usage limitations, allowing readers to assess whether this proactive agent is worth adding to their workflows.

11 additional sources reported this
OpenAI@OpenAIProductsAI Score83

OpenAI releases a Codex skill to help websites check their exposure to critical PHP vulnerabilities

OpenAI provides a Codex skill for critical PHP vulnerabilities, guiding users to run the github.com/openai/php2shell-checker repository to check whether their websites are affected. The context is OpenAI's Patch the Planet program, in which OpenAI models helped researchers discover critical PHP flaws. The fix involved major changes to the PHP engine and was disclosed after approximately two months of coordination. Websites using PHP can upgrade their versions and apply mitigation measures from security vendors.

Why it's recommended:The original article provides the official Codex skills and detection repositories, enabling website operators to check whether they are affected by the PHP vulnerabilities.

10/21 · Wednesday6 items

vLLM Official BlogProductsAI Score67

Ant Group and the vLLM community open-sourced the Humming quantized inference kernel library

Ant Group and the vLLM community open-source Humming, a general-purpose, high-performance kernel library for quantized inference, at github.com/vllm-project/humming.

Why it's recommended:The original article provides the kernel design for unified quantized representations plus JIT compilation, along with multiple benchmark figures, allowing readers to evaluate quantized-inference options across different GPUs.

Modal Official Engineering BlogProductsAI Score61

Modal launched Sticky Sessions: requests from the same session are always routed to the same container.

Modal released Sticky Sessions, which uses Modal Servers to route all requests from the same session to the same container. It can be created with the @modal.sessioned() decorator together with the SDK or HTTP API.

Why it's recommended:The article explains Sticky Sessions' routing, authorization, and autoscaling design, along with use cases such as voice agents, shared rooms, and games.

Fireworks AIProductsAI Score76

Meta's open-weight model Muse Spark Open launches on Fireworks' US serverless

Meta's open-weight model Muse Spark Open launches Day-0 serverless hosting on Fireworks. It uses an 816B-parameter MoE architecture, activates 42B parameters per token, and supports a 1M-token context window.

Why it's recommended:Fireworks provided specific scores for Muse Spark Open across multiple independent benchmarks in areas including law, finance, and medicine, as well as details on how it is made available, allowing readers to assess its suitability for regulated work.

Tibo@thsottiauxProductsAI Score65

The OpenAI API supports 12-hour prompt caching, while caches for Codex and Work are retained permanently

OpenAI's developer account announces that the API now offers optional 12-hour prompt caching, allowing continuously running agents to reuse instructions, tools, and context at lower cached-input rates even when requests are hours apart. Tibo adds that caches for signed-in users in Codex and Work are retained permanently, with guaranteed cache hits when returning to old conversations or resuming sessions hours later, and no additional penalty.

Why it's recommended:The author describes firsthand the 12-hour API cache and persistent caching strategies on Codex and Work, allowing readers to directly assess their impact on the cost and experience of long-running Agents.

Also reported by 1 other source
Databricks:BlogProductsAI Score62

Databricks released MemCurrent, a production-grade persistent-memory architecture for agents.

Databricks introduced MemCurrent, a persistent-memory architecture for production agents. It combines episodic and derived memory and reads, writes, and integrates memories through three access modes: plugins, tool calls, and an offline dreaming pipeline. It is already used in Agent Bricks Memory and Genie.

Why it's recommended:The article provides the full design, ablation results, and cost data, allowing readers to evaluate architecture choices and benefits for production-grade agent memory.

PyTorch:BlogProductsAI Score74

Google and Meta release TorchTPU, a native TPU backend for PyTorch.

Google and Meta jointly open-sourced TorchTPU, providing a native PyTorch backend for Google TPUs. Models can run out of the box by changing .to("cuda") to .to("tpu"), without requiring a JAX dependency.

Why it's recommended:The original article details the architecture, precision modes, and multiple training and inference cases, allowing readers to assess the cost of migrating PyTorch workflows to TPUs.

Also reported by 3 other sources

10/20 · Tuesday7 items

Aravind Srinivas@AravSrinivasProductsAI Score65

Perplexity Computer can now run entirely locally on Mac, without cloud inference

Perplexity CEO Aravind Srinivas announces that Perplexity Computer can now run entirely locally on Mac, without cloud inference or Wi-Fi. Both the model and agent run on the machine and can operate native Mac applications and local files, so sensitive data does not need to be sent to the cloud. It is now available in the Perplexity Mac app.

Why it's recommended:The Perplexity CEO personally announced that Perplexity Computer can run entirely locally, so sensitive data does not need to be uploaded to the cloud, making it relevant to people concerned about privacy and offline use.

Also reported by 2 other sources
Google Developers BlogProductsAI Score71

Google released native TorchTPU inference engines for vLLM and SGLang

In a PyTorch-related announcement, Google introduces native PyTorch inference engines for vLLM and SGLang based on TorchTPU, developed with Inferact and RadixArk respectively, allowing developers to deploy models on TPUs using familiar PyTorch tools.

Why it's recommended:The original article provides feature lists, public benchmark costs, and open-source entry points for two native PyTorch TPU inference engines, allowing readers to evaluate the feasibility of migrating an existing PyTorch service stack to TPU.

Also reported by 2 other sources
IBM Research:AIProductsAI Score60

llm-d 1.0 released: Kubernetes-native distributed inference service becomes production-ready

The open-source inference project llm-d, led by IBM, Red Hat, Google, and others, released llm-d 1.0 at PyTorch Conference North America. It is a Kubernetes-native distributed inference service framework that supports any model, any accelerator, and any cloud.

Why it's recommended:The original article provides specific capability changes and measured cost data for llm-d 1.0, allowing readers to assess whether building their own inference infrastructure is worthwhile.

Also reported by 1 other source
Cursor BlogProductsAI Score71

Cursor releases Agent SDK, enabling agents to be defined in code and deployed with one click

Cursor released Agent SDK, the framework it uses to build hosted agents such as Bugbot, Security Reviewer, and Rollouts. It is now fully available.

Why it's recommended:Cursor open-sourced the development framework it uses for hosted agents such as Bugbot, enabling readers to assess the feasibility of defining and deploying team agents with code.

ElevenLabs:BlogProductsAI Score73

ElevenLabs releases Eleven Realtime v1 Mini and v1 Max voice-agent models

ElevenLabs released the Speech to Speech-level Eleven Realtime models, including v1 Mini for low-cost natural conversation and v1 Max for complex, multistep tasks. Both support more than 90 languages and 17,000+ voices.

Why it's recommended:The original article provides benchmark data for the two model tiers and a migration method compatible with the OpenAI Realtime API, allowing readers to determine whether to replace their existing voice-agent solution.

WeChat Official Account: DeepSeekProductsAI Score75

DeepSeek fully open-sources its internal reasoning framework, DeepInfer

DeepSeek announces that it will fully open-source its internal reasoning framework DeepInfer, while also open-sourcing the production-grade Token encoding and decoding component deepseek-recipe.

Why it's recommended:DeepSeek will fully open-source its internal reasoning framework, providing repository addresses and compatible platforms so developers can optimize the cost of large-scale inference.

Also covered by 7 other sources
OpenRouter@OpenRouterProductsAI Score67

Inception's diffusion model Mercury 3 launches on OpenRouter

Inception's diffusion model Mercury 3 launches on OpenRouter, generating at over 1,000 tok/s and supporting a 1M context, four times that of Mercury 2.5. Inception reported that it surpasses GPT-6 Luna and Haiku 5.5 on IFBench, GPQA Diamond, and SWE-Bench Atlas; during the launch period, it is offered at half price, at $0.10/M for input and $0.35/M for output.

Why it's recommended:The original article provides the speed, context, benchmark comparisons, and launch discount pricing of Inception's diffusion model, which can be used to assess its cost-effectiveness relative to existing models.

10/19 · Monday2 items

PyTorch:BlogProductsAI Score67

AWS releases TorchNeuron: Trainium’s native PyTorch backend open-sourced

AWS announced TorchNeuron, an open-source native PyTorch backend for Trainium, which entered public beta on October 20 under the Apache 2.0 license at PyTorch Conference North America.

Why it's recommended:The original article provides TorchNeuron’s architecture, execution modes, and compatibility test data, allowing developers to assess the cost of migrating existing PyTorch workflows to Trainium.

Also reported by 3 other sources
Load more

Finished reading the news? Let’s talk about your business.

Leave your contact details, and I’ll get in touch with you personally.

WECHAT

Add me on WeChat

Press and hold or select the WeChat ID below to copy it.