Skip to content

All news

16 today

Apr 22Wednesday

QbitAI · WeChat

SenseAuto's Sage with 3B active params claims to beat GPT-5.4 and Opus 4.6 in cars

SenseAuto released Sage, an in-car multimodal edge model with 32B total params and 3B active params, and says it scored 94% on PinchBench, above Claude Opus 4.6 at 93.3% and GPT-5.4 at 90.5%. The post says Sage runs on Nvidia OrinX with about 0.5s TTFT, 0.03s TPOT, and 80 tok/s throughput; its SCOUT training method cuts GPU hours by about 60%, and ERL raises complex-task completion by 20%. The key point is not the headline race but whether a 3B-active model can sustain multi-step tool use on device.

Why it matters: SenseAuto's Sage is an on-car multimodal model with 32B total parameters that activates 3B per pass, scoring 94% on PinchBench ahead of Claude Opus 4.6 and GPT-5.4, with OrinX latency and throughput numbers.

Synced · WeChat

Honor preinstalls YOYO Claw on MagicBook, calling it the world's first "agent laptop"

Honor said it preinstalls its YOYO Claw on MagicBook and claims 50% lower total token use than an OpenClaw setup. The post says it ships with 5 primary agents and 23 sub-agents, plus local processing, second-step confirmation, and kernel-level encryption. The practical angle is packaging agents as a device default, but the post does not disclose model names, hardware specs, pricing, or launch timing.

Why it matters: Honor preloads YOYO agents into MagicBook, so agents ship as a default laptop capability rather than something users deploy themselves.

Latent Space

OpenAI launches GPT-Image-2

OpenAI shipped GPT-Image-2 in ChatGPT, Codex, and the API. It has thinking and non-thinking variants, with stronger text, layout, editing, multilingual output, and QR codes. Arena ranks it first on 3 Image Arena boards, with 1512 Elo in text-to-image and a +242 lead.

Why it matters: OpenAI put GPT-Image-2 into ChatGPT, Codex and the API, with better text rendering, layout editing and multilingual output for slides, infographics and UI mockups.

TechCrunch · AI

Meta will record employees’ keystrokes and use it to train its AI models

Meta says a new internal tool converts employees’ mouse movements and button clicks into training data for its AI models. The headline mentions keystrokes, but the post only discloses mouse and click signals, not collection scope, consent, or retention terms. The real issue is the missing internal data-governance detail.

Why it matters: Meta is collecting real mouse and keyboard data from employees to train AI on everyday computer tasks, making the method and use of that data the main point.

X · @dotey

Anthropic quietly removed Claude Code from the $20 Pro plan on its pricing page without an announcement

Anthropic was spotted removing Claude Code from the $20 Pro plan on its pricing comparison page without an announcement. The snippet says help docs also removed the inclusion, while the Claude Code product page and support bot still say it is included, and some Pro users report access still works; the post does not disclose Anthropic’s formal explanation or effective date. The key issue is price floor: if confirmed, entry cost for Claude Code rises from $20 to $100 per month.

Why it matters: Anthropic's own pages disagree on whether Pro still includes Claude Code, and no explanation or effective date has been given.

Hacker News front page

Anthropic removes Claude Code from the $20/month Pro subscription for new users

Anthropic was reported to remove Claude Code from the $20/month Pro plan for new users, while saying existing Pro and Max subscribers are unaffected. The cited evidence: an April 10 archived help page said “Pro or Max plan,” the current page says “Max plan,” and Amol Avasare said this is a test on about 2% of new prosumer signups. The key issue is whether pricing shifts fully to Max or API billing; the post does not disclose retroactive scope or a final rollout timeline.

Why it matters: Anthropic pulled Claude Code from the $20-a-month Pro plan and left it only on Max, so new users no longer get the coding tool while existing users are unaffected for now.

X · @dotey

OpenAI launches ChatGPT Images 2.0, available to all ChatGPT and Codex users starting today

OpenAI made ChatGPT Images 2.0 available today to all ChatGPT and Codex users, and also opened the gpt-image-2 API. The RSS snippet says it supports up to 2K output, aspect ratios from 3:1 to 1:3, and more reliable non-English text rendering. In thinking mode, it can search the web, generate multiple styles, and self-check outputs; that tier is limited to Plus, Pro, and Business, with Enterprise not yet available.

Why it matters: ChatGPT image generation reaches 2.0 with free-tier access, steadier complex layouts and multilingual text, up to 2K output and aspect ratios from 3:1 to 1:3.

X · @OpenAI

Introducing ChatGPT Images 2.0

OpenAI introduced ChatGPT Images 2.0 as an image model for complex visual tasks and directly usable visuals. The RSS snippet cites sharper editing, richer layouts, and “thinking-level intelligence,” but the post does not disclose model size, pricing, latency, or rollout scope.

Why it matters: OpenAI's ChatGPT Images 2.0 announcement claims finer edits, richer layouts and thinking-level intelligence, but gives no model size, pricing or rollout details.

Bloomberg Technology

OpenAI unveils new image model that is better at charts and diagrams

OpenAI released an update to its image generation software to produce more accurate, complex charts and scientific diagrams. The RSS snippet does not disclose the model name, launch timing, pricing, benchmarks, or technical method. The real signal is a push into professional use cases, not generic image quality.

Why it matters: OpenAI's image model update targets accuracy on complex charts and scientific diagrams rather than general image quality.

The Verge · AI

OpenAI’s updated image generator can now pull information from the web

OpenAI said ChatGPT Images 2.0 can pull information from the web when a thinking model is selected, helping generate multiple images from one prompt. It runs on GPT Image 2 and is available to ChatGPT Plus, Pro, Business, and Enterprise users; the post does not disclose rollout timing, usage limits, or pricing changes. The key shift is web-grounded multi-image generation, not just image quality.

Why it matters: ChatGPT Images 2.0 can now search the web before generating a set of images, available to Plus, Pro, Business and Enterprise users.

X · @dotey

Google splits Gemini Deep Research into Deep Research and Deep Research Max

Google split Gemini Deep Research into Deep Research and Deep Research Max, with public preview starting today in paid Gemini API tiers. Both run on Gemini 3.1 Pro; one targets speed and cost, while Max runs longer with more compute and repeated search and reasoning. The update adds MCP support for sources such as FactSet, S&P, and PitchBook, plus files, code execution, and File Search; the post does not disclose pricing.

Why it matters: Google splits Gemini Deep Research into a standard tier built for speed and cost and a Max tier that spends more compute on repeated search and reasoning.

The Verge · AI

Celebrities will be able to find and request removal of AI deepfakes on YouTube

YouTube is expanding its AI deepfake monitoring tool to Hollywood celebrities, letting enrolled public figures find impersonation videos and request takedowns. Flags are reviewed under YouTube's privacy policy, so not every request is approved. The tool was tested with creators last fall and expanded to politicians and journalists in March; the post does not disclose rollout size or timing.

Why it matters: YouTube now lets celebrities register for automatic scanning of impersonation videos and file takedown requests, after earlier rollouts to creators, politicians and journalists.

Apr 21Tuesday

QbitAI · WeChat

Mystery model Elephant: 100B parameters reaches same-scale SOTA with high token efficiency

Ant Group's Inclusion AI team is identified as the maker of Elephant, a 100B-parameter model with 256K context and 32K output shown on OpenRouter. The post reports tests on bug fixing, summarizing a 3,000-word meeting note, and a light agent loop, plus AI BENCHY figures of about 2,500 output tokens, about 1 second average latency, and 9.6/10 consistency; the post does not disclose training details, pricing, or an official model card.

Why it matters: Ant Digital's Inclusion AI team put its 10-billion-parameter Elephant model on OpenRouter, and the piece tests it on bug fixes, a 3,000-word meeting summary and a light agent workflow.

Hacker News front page

CrabTrap: An LLM-as-a-judge HTTP proxy to secure agents in production

Brex open-sourced CrabTrap, an HTTP proxy that intercepts every agent request and allows or blocks it against a policy in real time. The page shows a dual path of static rules plus an LLM judge, and logs whether each decision came from rule matching or model judgment; the post does not disclose the model, latency overhead, or error rates.

Why it matters: Brex open-sourced CrabTrap, an HTTP proxy that sits between agents and external services, matching requests against static rules first and sending the rest to an LLM judge.

Ben's Bites

That's My Designer - Claude

Anthropic added a Design tab to Claude that asks 5-10 interactive questions, then builds wireframes or high-fidelity prototypes. The post says image-to-design works well; in research preview it has separate limits, and the $20 plan appears to allow only 2-3 large generations per week. The sharper point is usability: the author says Claude Cowork depends on connectors and plugins that average users may not find.

Why it matters: Anthropic added a design tab to Claude that turns 5 to 10 rounds of questions into wireframes or high-fidelity prototypes, with usage limits disclosed for the research preview.

Synced · WeChat

Sergey Brin revives founder mode? Google forms a strike team to focus on AI coding

Google has formed an AI coding strike team led by Sebastian Borgeaud, with Sergey Brin and Koray Kavukcuoglu directly involved, to improve long-context coding and internal code automation. The pressure signal cited is that Google said about 50% of its code is written by coding agents and reviewed by engineers, while Anthropic staff claimed 100% code use by Claude Code and Opus 4.5; the post does not disclose team size, launch timing, or the exact Google model version. The key issue is whether Google can turn private codebase training into stronger public models.

Why it matters: Google formed an AI coding strike team led by a former DeepMind researcher, with Brin and Koray Kavukcuoglu directly involved, targeting long-context coding and internal code automation.

OpenAI News

Introducing ChatGPT Images 2.0

OpenAI introduced ChatGPT Images 2.0 as a new image generation model, highlighting better text rendering, multilingual support, and visual reasoning. The RSS snippet names only these three upgrades; the post does not disclose architecture, resolution, pricing, latency, or availability. What matters is whether text fidelity and multilingual consistency improve in real use; for now, only headline-level details are disclosed.

Why it matters: OpenAI released ChatGPT Images 2.0, claiming gains in text accuracy, multilingual layout and visual reasoning on complex prompts, with sample posters, comics and magazine pages.

Xinzhiyuan · WeChat

OpenAI launches Chronicle research preview for Codex with screen context

OpenAI launched Chronicle research preview for Codex on April 21. It is limited to ChatGPT Pro users on Mac and reads recent screen context to reduce repeated background prompts. OpenAI says data is “primarily processed locally,” but the post says some cases use cloud help; The Next Web reports screenshots are uploaded and local memories are unencrypted, while upload share and retention time are not disclosed.

Why it matters: OpenAI launched a research preview of Codex's Chronicle, limited to ChatGPT Pro users on Mac, which reads recent screen content so users stop re-explaining context.

Xinzhiyuan · WeChat

Huawei launches Pura X Max with debut Xiaoyi companion AI

Huawei launched Pura X Max on April 20 and debuted Xiaoyi companion AI on HarmonyOS 6.1. The post says it can be invoked by double-tapping the nav bar or voice, read screen content with consent, collect tasks across apps into Calendar, and connect with Amap and Didi. The key point is system-level cross-app access and persistent side-panel UX; the post does not disclose price, model specs, or coverage.

Why it matters: Huawei's Pura X Max shows system-level AI reading the screen with consent and moving tasks across apps, a concrete path to cross-app agents.

X · @dotey

GitHub paused new sign-ups for Copilot Pro, Pro+, and Student on April 20

GitHub paused new sign-ups for Copilot Pro, Pro+, and Student on April 20, leaving only Copilot Free open to new users. The post says Pro+ now has more than 5x Pro usage, Claude Opus 4.7 is limited to Pro+, and users can request cancellation with a full April refund from Apr 20 to May 20. What matters is the price stayed fixed while access, quotas, and model tiers tightened first.

Why it matters: GitHub paused new signups for Copilot Pro, Pro+ and the student tier, pulled Claude Opus from the Pro plan, and raised Pro+ quotas to more than five times Pro.

Hacker News front page

OpenAI ad partner now selling ChatGPT ad placements based on "prompt relevance"

The headline says an OpenAI ad partner is already selling ChatGPT ad placements using “prompt relevance” for targeting. The link points to an Adweek report on StackAdapt, but only an RSS snippet is provided. The post does not disclose placement, auction logic, pricing, reach, or launch timing; the key issue is whether chat context is becoming ad inventory.

Why it matters: Ad platform StackAdapt is recruiting advertisers to test ChatGPT ad slots matched by prompt relevance, with CPMs of $15 to $60 and a $50,000 pilot minimum.

X · @claudeai

In Cowork, Claude can now build live artifacts: dashboards and trackers connected to your apps and files

Claude added live artifact building in Cowork, letting users create dashboards and trackers tied to apps and files. Opening an artifact refreshes current data; the post does not disclose supported apps, file sources, or permission controls.

Why it matters: Claude in Cowork can now build dashboards and trackers that pull fresh data from connected apps and files each time they open, unlike static screenshots.

Bloomberg Technology

Google to Release New AI Chips, Challenging Nvidia | Bloomberg Tech 4/20/2026

Google plans to release new AI chips focused on inference, directly challenging Nvidia. The RSS snippet confirms the inference focus, but the post does not disclose launch timing, model names, performance, pricing, or customers. The real signal is rising competition on inference silicon supply, not the show's other rocket or IPO items.

Why it matters: Google is preparing a new AI chip built for inference, putting it in direct competition with Nvidia at the same stage of the stack.

X · @dotey

OpenAI adds Chronicle to Codex, letting it read screen context

OpenAI added Chronicle to Codex and is rolling it out to ChatGPT Pro users on macOS; it uses periodic screenshots, OCR, and tool detection to turn recent screen activity into memory. The memory is stored as plain Markdown in ~/.codex/memories_extensions/chronicle, and the EU, UK, and Switzerland are excluded; OpenAI says screenshots are uploaded for processing, deleted afterward, and not used for training. The part to watch is risk: the background agent can burn rate limits, local plain-text files widen exposure, and OpenAI warns it amplifies prompt-injection from malicious webpages.

Why it matters: OpenAI is testing Chronicle in the macOS ChatGPT Pro app, letting Codex take timed screenshots and turn them into callable local memory so users can say "this error" without re-explaining context.

Bloomberg Technology

Google to Release New Inference-Focused Chips

Google plans to announce a new generation of custom TPUs this week, aimed at AI inference workloads. The RSS snippet confirms only the timing and chip focus; model names, performance, power, and pricing are not disclosed. Watch inference cost and supply, not the headline alone.

Why it matters: Google is set to release a new custom TPU aimed at AI inference this week, and with model, performance and price undisclosed, the launch itself signals where compute supply is heading.

The Verge · AI

Fortnite developers can make AI characters now — just don’t try to date them

Epic Games is rolling out a “conversations” tool for Fortnite creators, turning island NPCs into AI characters that can talk with players in unscripted ways. The snippet says creators define persona, knowledge, behavior, and voice with prompts; the title says don’t try to date them, but the post does not disclose the exact guardrails or moderation system.

Why it matters: Epic launched a dialogue tool in Fortnite, letting creators set an island NPC's persona, knowledge, behavior and voice through prompts so players can talk off-script.

Latent Space

Training Transformers to Address the 95% Failure Rate in Cancer Trials — Noetik

Noetik uses TARIO-2 to predict tumor spatial transcriptomics, targeting a 95% cancer-trial failure rate. GSK signed a $50M technology deal, and TARIO-2 predicts a ~19,000-gene spatial map from routine H&E assays. The key issue is patient-tumor-treatment matching, not the claim that AI cures cancer.

Why it matters: Noetik trained an autoregressive Transformer to predict spatial expression of about 19,000 genes from routine H&E slides, tracing cancer trials' high failure rate to patient, tumor and drug pairing.

Apr 20Monday

X · @Yuchenj_UW

Kimi K2.6 is open-source

Kimi K2.6 is now open source, and the RSS snippet says it scored 58.6 on SWE-Bench Pro. The snippet also says it beat GPT-5.4 xhigh and Claude Opus 4.6 max effort. What matters is reproducibility; the post does not disclose weights, license, or eval setup.

Why it matters: Kimi K2.6 is open sourced and scores 58.6 on SWE-Bench Pro for code bug fixing, above GPT-5.4 and Claude Opus 4.6 in the tweet, adding a reference point for open versus closed code ability.

Synced · WeChat

In the first year of “deployment mode,” AgiBot expanded its rollout plans to seven solutions

AgiBot said at its April 17 Shanghai event that it released 4 robots, 6 AI models, and 7 standardized deployment solutions, and framed 2026 as the first year of embodied AI “deployment mode.” The post cites concrete metrics: Expedition A3 runs 8-10 hours, WITA Omni 1.0 targets sub-500ms interaction latency, and BFM was trained on 100 million-plus frames and 700 hours of motion-capture data; it also claims 5,100-plus shipments and 39% share in 2025, with the 10,000th robot rolling off in March 2026. The real point for practitioners is repeatable delivery rather than launch volume: the post lists 7 scenarios from 3C line loading to patrol, but independent validation details are not disclosed.

Why it matters: Zhiyuan unveiled four robots, six AI models and seven standardized deployment packages in Shanghai, with concrete battery, interaction-latency and training-data figures, framing 2026 as the first year of deployment.

Bloomberg Technology

China’s Netflix iQiyi Goes All-In on AI Content in Big Overhaul

iQiyi has begun the biggest overhaul in its 16-year history, aiming for AI to generate a sizable share of films and shows from scratch someday soon. The RSS snippet gives only that direction; the post does not disclose models, spending, content share, or launch timing. The real signal is the scale of the reorganization, not the slogan.

Why it matters: iQiyi's biggest reorganization in 16 years folds AI-generated content into slate planning, an organizational shift in how a major long-video platform makes shows.

Apr 19Sunday

Synced · WeChat

Amap debuts an autonomous embodied robot at the Yizhuang Marathon and showcases guide-assistance

Amap showed its quadruped robot Tutu at the 2026 Yizhuang humanoid half marathon, claiming it completed a guide-assistance obstacle task in an open environment without preset routes or teleoperation. The post says its ABot stack includes ABot-N0, which reached SOTA on 7 navigation benchmarks with 88.3% on SocNav, and ABot-M0, which scored 80.5% on Libero-Plus. The key point is the integrated stack across navigation, manipulation, world modeling, and closed-loop correction; the post does not disclose guide-task test scope, commercialization timing, or safety incident data.

Why it matters: A quadruped robot navigated open roads with no preset route or remote control, a rare public test of real-world navigation.

QbitAI · WeChat

Amap unveiled ABot, its first full-stack embodied AI stack for AGI, and claimed 15 SOTA results

Amap unveiled embodied AI stack ABot and claimed SOTA on 15 metrics. The post says ABot-3DGS builds 10k-scale 3D scenes from centimeter-level map data, while ABot-PhysWorld uses a 14B DiT and 3M real manipulation videos. What matters is the interactive world model and VLA loop; the post does not disclose the 15 benchmarks, exact metrics, or the open-source timeline and scope.

Why it matters: Amap published its full-stack embodied intelligence system ABot, whose ABot-PhysWorld trains a 14B-parameter diffusion model on 3 million real robot-arm manipulation videos so robots simulate an action before executing it.

Xinzhiyuan · WeChat

Amap unveiled ABot-Claw and its quadruped robot Tutu at the Yizhuang Half Marathon

Amap unveiled the ABot-Claw agent system and the quadruped robot Tutu, claiming an autonomous guide-dog demo in the 2026 Yizhuang robot half marathon. The post gives three concrete numbers: ABot-M0 reached 80.5% on Libero-Plus, nearly 30% above Pi0; ABot-N0 hit SOTA on 7 navigation benchmarks; the open UniACT dataset contains 6 million trajectories and 9,500+ hours. What matters is Map as Memory, cloud-edge control, and closed-loop self-correction; the post does not disclose race ranking, pricing, or launch timing.

Why it matters: At the Yizhuang half marathon Amap showed ABot-Claw and the quadruped Tutu guiding a blind runner autonomously, with ABot-M0 scoring 80.5% on Libero-Plus, nearly 30 points above Pi0, and navigation at SOTA on 7 benchmarks.

Apr 18Saturday

Synced · WeChat

Claude Design enters research preview for generating mockups, prototypes, and slides

Anthropic launched Claude Design in research preview for Claude Pro, Max, Team, and Enterprise users, covering mockups, prototypes, slides, and one-pagers. Powered by Claude Opus 4.7, it can ingest codebases, images, DOCX, PPTX, XLSX, and web captures, then export to Canva, PDF, PPTX, and HTML; the headline cites Figma and Adobe stock drops, but the post does not disclose the moves. The real signal is the workflow link from design system ingestion to handoff into Claude Code.

Why it matters: Anthropic launched a Claude Design research preview on paid Claude accounts that reads codebases, images and several document types, and exports to Canva, PDF, PPTX and HTML.

Xinzhiyuan · WeChat

Claude Opus 4.7 splits users 48 hours after launch: benchmark lead, reasoning tests drop

Anthropic's Claude Opus 4.7 drew split reactions within 48 hours: Artificial Analysis scored it at 57, tied for No.1, while NYT Connections Extended fell from 94.7% on 4.6 to 41.0%. The post says a new tokenizer raises token usage to 1.0-1.35x on the same text, and old thinking parameters can return 400 errors; Anthropic also cites a 1753 Elo GDPval-AA score, 79 points above No.2. The real issue is migration cost and capability trade-offs, not a single leaderboard.

Why it matters: Two days after launch, Anthropic's Claude Opus 4.7 ties for first on aggregate scores, but its NYT Connections Extended reasoning accuracy fell to 41.0% from 94.7% on 4.6.

Xinzhiyuan · WeChat

Bilibili debate: Hermes responds to plagiarism claims for the first time, as MiniMax moves early on Harness

MiniMax says its M2.7 model now handles 30%-50% of daily workflows in its RL team, ran over 100 self-optimization loops, and improved evals by 30%. The post also says Hermes Agent grew from 2B to nearly 300B daily tokens, while M2.7 exceeds 25B daily tokens on OpenRouter; Hermes lead Tommy Eastman denied copying EvoMap in a livestream. The real signal is Harness: the post cites 20-40ms or 80ms sandbox startup and 15k to 600k instances per minute, showing competition is shifting from benchmark scores to agent execution infrastructure.

Why it matters: MiniMax says M2.7 now handles 30% to 50% of its reinforcement learning team's daily work after over 100 self-optimization rounds, while Hermes Agent calls grew from 2 billion to nearly 300 billion tokens a day.

Hacker News front page

Show HN: AI Subroutines – Run automation scripts inside your browser tab

rtrvr.ai introduced AI Subroutines, which turn a recorded browser task into a callable tool and replay it at zero token cost and zero LLM inference delay. The script runs inside the active tab, reusing auth, CSRF, TLS sessions, and signed headers; recording trims about 300 requests to about 5 and falls back to DOM-only when GraphQL operation IDs are volatile. The part to watch is batching: one LLM call can assign parameters for a 500-row sheet and launch 500 subroutines.

Why it matters: rtrvr.ai records a browser action into a reusable script that replays in the current page, with login state and request signing handled by the browser.

X · @dotey

Anthropic launches Claude Design, a conversational design generation product

Anthropic released Claude Design in research preview and is rolling it out to Pro, Max, Team, and Enterprise subscribers. Powered by Claude Opus 4.7, it can start from text, images, docs, or web clips, then iterate via chat, comments, direct edits, and sliders. On first use it reads a team's codebase and design files to build a design system; outputs export to Canva, PDF, PPTX, or standalone HTML, with one-click handoff to Claude Code.

Why it matters: Anthropic's Claude Design turns conversation into design files, and on first use reads the codebase and design files to build a full system of brand colors, fonts and components.

Apr 17Friday

X · @op7418

The previously leaked Claude design tool is now live

Claude has launched a design tool that can generate web pages, app prototypes, and PPTs, based on an RSS snippet. The snippet confirms PPT export and export to Canvas; the product name, pricing, plan access, and regional limits are not disclosed. The key shift is from chat output to design artifact generation and export.

Why it matters: Claude now has a design tool that generates web pages, app prototypes and slides, exports to PPT, and hands files to Canvas for further editing.

Hacker News front page

Introducing Claude Design by Anthropic Labs

Anthropic launched Claude Design on April 17, 2026, in research preview for Claude Pro, Max, Team, and Enterprise subscribers. Powered by Claude Opus 4.7, it generates designs from text, images, DOCX, PPTX, XLSX, and codebases, and exports to Canva, PDF, PPTX, or HTML. The key detail is its one-step handoff bundle to Claude Code; the post does not disclose standalone pricing beyond existing plan limits.

Why it matters: Anthropic opened a research preview of Claude Design, where Pro, Max, Team and Enterprise users turn text, images, documents or a codebase into designs and prototypes.