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 11/19 03:51 (UTC+8) · refreshed hourly

10/08 · Thursday26 items

Claude Code:GitHub ReleasesProductsAI Score60

Claude Code v2.1.293 Released: Adds Claude Haiku 5.5 and Fixes Numerous Issues

Claude Code released v2.1.293, adding Claude Haiku 5.5 (claude-haiku-5-5) as the default Haiku model for the Anthropic API, with support for a 1M context window and pricing of $0.10/$0.50 per million tokens ($0.50/$2.50 for prompts over 100K).

Why it's recommended:The release notes list a new default Haiku model, expanded scripting interfaces, and numerous stability fixes, helping readers decide whether to upgrade and assess the impact on existing workflows.

ClaudeDevs@ClaudeDevsProductsAI Score70

Anthropic Introduces Monthly Platform API Credits for Claude Max and Team Plans

Anthropic is introducing monthly Claude Platform API credits for Claude Max and Team plans: $100 for Max 5x, $200 for Max 20x, and up to $500 for Team, with sharing supported. The credits apply to any model, including Haiku 5.5, and can be used in your own code or third-party harnesses.

Why it's recommended:The original article provides the monthly API allowance amounts and applicable scopes for the Max and Team plans, enabling readers to determine whether adjusting their current subscription is worthwhile.

Also reported by 1 other source
Claude@claudeaiProductsAI Score65

Anthropic Cuts Claude Sonnet 5.5 Cache Read Pricing in Half to $0.10 per Million Tokens

Anthropic announced that it would halve the cache read price for Claude Sonnet 5.5 to $0.10 per million tokens. The company says this reduces the operating cost of Sonnet 5.5 by approximately 20% for most long-running tasks.

Why it's recommended:The official announcement states that cache read prices have been halved and gives the cost reduction for long-running tasks, allowing users to directly evaluate the cost of using Claude Sonnet 5.5.

Also reported by 1 other source
Anthropic:NewsroomModelsAI Score78

Anthropic Releases Claude Haiku 5.5, Reducing Operating Costs by Approximately 75% on Average

Anthropic released Claude Haiku 5.5, positioning it as the cheapest and fastest small model to date. It is suited to high-throughput tasks such as summarization, compression, database queries, and classification, and can work alongside Opus 5.5 and Sonnet 5.5 as a coding sub-agent.

Why it's recommended:The official announcement includes pricing, benchmark results, and changes to subscription API allowances, helping readers assess where small models fit in low-cost agent tasks.

Reported by 12 other sources
Stanford HAI NewsPapersAI Score61

Stanford HAI Study: To Get Employees to Embrace AI, Don't Package It as a Productivity Tool

Stanford HAI reports on Karunakaran and colleagues’ study of AI deployment at law firms, advertising agencies, and IT services companies. They found that employees were reluctant to use AI when it was presented as a productivity tool, but were much more willing when it was framed as a way to enrich their work. In a controlled observation at a law firm, the department that introduced the AI tool LawBot to enrich work saw 58% more usage and 70% more experimentation with new uses than the department focused on efficiency. The study also emphasizes the importance of supporting training, time to explore, and adjustments to performance metrics.

Why it's recommended:The study offers a practical management approach: framing AI as a way to enrich work, while providing training and time, does more to encourage employees to actually use it than emphasizing efficiency.

Liquid AI models and engineering blogModelsAI Score61

Liquid AI releases open-source decision models d1-3B and d1-omni-600M

Liquid AI has released the open-weight decision models d1-3B and d1-omni-600M, now available on Hugging Face.

Why it's recommended:The official release provides the model architectures, training details, and latency data across hardware platforms, allowing readers to assess the trade-offs of deploying small-parameter decision models at the edge.

Also reported by 3 other sources
Unsloth AI@UnslothAITutorialsAI Score66

Unsloth Open-Source Tutorial: Train a Qwen3.5 0.8B Decision Model Locally, Improving Accuracy from 20.7% to 74.3%

Unsloth released a tutorial and open-source repository that can fine-tune LLMs such as Qwen3.8 and Gemma 4 into decision models that output probabilities for options. Qwen3.5 0.8B achieved a combined accuracy increase from 20.7% to 74.3% on 3 decision-making benchmarks, using only 4GB of VRAM.

Why it's recommended:The original article provides a complete fine-tuning method and an accuracy comparison before and after. Readers can reproduce the process of turning an LLM into a decision model on a GPU with 4GB of VRAM.

Aravind Srinivas@AravSrinivasModelsAI Score65

Perplexity Open-Sources the pplx-embed-v2-late Multimodal Late-Interaction Embedding Model (9B and 0.6B)

Perplexity open-sourced pplx-embed-v2-late, two late-interaction multivector embedding models for text and images, sized at 9B and 0.6B. They share the same embedding space, and their weights are available on Hugging Face. The 9B model can index multimodal data, while the 0.6B model can perform on-device queries and retrieve PDF pages without OCR; the models scored 92.4% on MADQA and 64% on BrowseComp+.

Why it's recommended:The original article provides two embedding-model sizes, explains how to use the shared space, and gives benchmark scores, allowing readers to assess viable approaches for multimodal retrieval and edge-side querying.

Also reported by 2 other sources
Microsoft Research BlogPapersAI Score69

Microsoft Research Asia Open-Sources Agent Lightning v1.0: A Real Harness Agent RL Training Framework in 3,500 Lines of Code

Microsoft Research Asia proposed the Harnessed Agentic RL training paradigm and open-sourced the reconstructed Agent Lightning v1.0, allowing the same agent harness used in deployment to participate directly in reinforcement learning without rewriting the agent within the training framework.

Why it's recommended:The framework lets the real agent harness used in deployment participate directly in RL training, without rewriting agent code, and provides end-to-end coding-agent training results for reference.

Also reported by 1 other source
Claude:YouTubeProductsAI Score66

Claude September recap: Chat and Cowork merge, and the Claude 5.5 model family launches

Claude released a September update recap: Chat and Cowork have been merged into a single Claude, workflows can run in the cloud and continue after you close your laptop. Claude Docs supports teams and Claude editing collaboratively in the same document. The newly launched Claude 5.5 series offers Opus 5.5 for heavy tasks and Sonnet 5.5 for quick edits, with /slides, /docs, and /designs enabling direct generation in the corresponding formats.

Why it's recommended:The official summary covers the key September updates, enabling readers to quickly understand the new features following the merger of Chat and Cowork and the division of responsibilities between the two models.

10/07 · Wednesday16 items

LlamaIndex: Products, Engineering, and EvaluationProductsAI Score64

LlamaIndex releases OpenDocRouter, an API for using multiple document-parsing models

LlamaIndex launches OpenDocRouter, a platform that parses documents (PDF, PNG, JPEG, or URL) into markdown through a unified API. It connects to 5 frontier models, including Claude Opus 5.5, Gemini 3 Flash, and GPT-6 Luna, and 5 open-source models, including MinerU2.5-Pro and PaddleOCR-VL-1.6.

Why it's recommended:The original article provides comparisons of the models' quality on ParseBench and cost per 1,000 pages, along with details on unified layout output and token-based billing, making it easier for developers to choose between them.

a16z:NewsOpinionsAI Score75

a16z Explains Why Texas Has Data Centers Waiting in Line for Power

a16z's Ryan McEntush analyzed why the Texas grid suspended approvals for data centers: the interconnection queue surged from 63 GW at the end of 2024 to 474 GW in June this year, with data centers accounting for about 90%; developers submitted large numbers of speculative applications and engaged insufficiently with communities.

Why it's recommended:The original article breaks down the causes of data center power restrictions in Texas through three threads—grid interconnection queues, behind-the-meter power, and flexibility—helping readers understand the true structure of the AI power bottleneck.

Google@GoogleTutorialsAI Score68

Google opens the SynthID Detector portal, which can detect whether images, videos, and audio were generated by AI

Google describes how to check whether media was generated by AI: visit https://synthid.com and upload an image, video, or audio file. The portal scans it for SynthID watermarks from Google or its partners.

Why it's recommended:The official source provides a specific tool and steps for detecting AI-generated media, so readers can follow the process to upload a file and check whether it contains a SynthID watermark.

Also covered by 7 other sources
Hugging Face Community BlogPapersAI Score75

Fine-tuned Nemotron models reach gold-medal level in IOI 2026 and IMO 2026

Using Nemotron 3, the NVIDIA Nemotron team built systems that achieved gold-medal-level performance at IOI 2026 and IMO 2026 through SFT, RL, and feedback-driven reasoning.

Why it's recommended:The article provides specific scores, training recipes, and open resources for gold-medal-level results at IOI and IMO, allowing readers to reuse its approach to combining models and reasoning systems.

Arena.ai@arenaModelsAI Score68

Mistral Large 4 enters Code Arena: WebDev ranks 45th with a score of 1534

Arena announced that Mistral Large 4 has joined Code Arena: WebDev, ranking 45th with a score of 1534, 304 points higher than Mistral Large 3 (ranked 130th) and 271 points higher than Mistral Medium 3.5.

Why it's recommended:Arena's test results provide comparisons of Mistral Large 4's coding score, ranking, and pricing, helping readers assess its value for money.

PromptArmor:Threat IntelligenceTutorialsAI Score60

PromptArmor analyzes WebMCP’s mechanisms and major security risks

PromptArmor published an analysis of WebMCP threat intelligence, explaining how WebMCP enables websites to provide tools to browsing agents and highlighting new risks such as governance, prompt injection, web security, misconfiguration, and operational drift.

Why it's recommended:From a frontline threat intelligence perspective, the author breaks down WebMCP’s mechanisms and emerging risks such as governance and prompt injection, enabling readers to assess their own vendor exposure.

Cursor:ChangelogProductsAI Score63

Cursor's iOS app adds remote control of local agents

Cursor has launched local agent remote control in its iOS app, allowing users to view and reply to agents running on their computers. It is enabled by default for all users except enterprise organizations. After logging in, computers associated with the account appear automatically, and the feature can be used once pairing is approved in the desktop app. Agents still run locally, so the computer must remain powered on and connected to the internet. An option to prevent sleep can be enabled in Settings, and enterprise administrators can enable this feature in Org settings.

Why it's recommended:The original text explains how to enable remote control, the pairing process, and the requirement that the computer remain powered on, allowing readers to assess whether it suits their own workflows.

Also reported by 1 other source
OpenAI: Website updatesPapersAI Score71

OpenAI releases mathematical research results produced by an internal frontier model

OpenAI releases a collection of new mathematical results produced by its internal frontier models. The results are publicly available as GitHub repositories, together with paper revisions and citation agreements. Many of the proofs have been formalized in Lean for computer verification.

Why it's recommended:The original article explains details such as how the results were released, Lean formalization, and computational costs, helping readers understand how AI-generated mathematical results are made available to the mathematics community.

Reported by 9 additional sources
Hacker News: Popular AI postsIndustryAI Score77

Mark Cuban criticizes Facebook for allowing AI deepfake scam ads; Meta internal documents say they generated approximately $16 billion in revenue

On the Pivot podcast, Mark Cuban said Meta knew that deepfake scam ads using AI-generated likenesses were running on its platform, such as ads selling a diabetes miracle cure for $294, and called Facebook the worst platform.

Why it's recommended:The content provides figures from Meta internal documents on the scale of risky advertisements and the review thresholds, allowing readers to compare them with the company’s response and assess the key points of contention surrounding the platform’s fraud governance.

GitHub BlogIndustryAI Score67

GitHub Rebuilds Git Infrastructure to Support Agent-Scale Development

GitHub announced that it is rebuilding its Git infrastructure to support the high-concurrency read and write loads created by agent-scale development. In August 2026, GitHub’s monthly Git event volume reached 473.3 billion, more than doubling year over year. In September, agents and developers generated 7.38 billion commits, more than five times the figure from a year earlier, while pushes grew 4.9-fold year over year.

Why it's recommended:GitHub engineers explain firsthand the motivations and design principles behind rebuilding Git infrastructure for agent-scale development, including internal benchmark data showing up to 35× higher write throughput.

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.