source&pool
A daily wire of long-form journalism, video, and discourse — filed, tagged, and laid out flat.
VOL. I·NO. 01
THURSDAY, SEPTEMBER 24, 2026
Hacker News3444X 主题热门3354CNBC59YahooFinance57aihot539to5Mac46MacRumors46IGN38Verge37Kotaku32TechCrunch25AndroidAuthority23Gematsu19NintendoLife19Engadget179to5Google15BusinessInsider15WarhammerCommunity15Eurogamer14USAToday14Guardian13ArsTechnica12Fortune11FoxBusiness11Polygon11PushSquare11Wccftech11Investor'sBusinessDaily10NPR10ABC9Gizmodo9NBC9Notebookcheck9SeekingAlpha9TechPowerUp9CBS8CNN8MotleyFool8GSMArena8NintendoEverything8VideoCardz8VideoGamesChronicle8AppleInsider7CNET7CoinDesk7Fox7PureXbox7Yahoo7BleepingComputer6Mashable6NewYorkPost6Conversation6AlJazeera5bgr5TechSpot5Tom'sGuide5WIRED5Aftermath4AndroidPolice4Deadline4Motor14XBOXWire4PetaPixel4PlayStationLifeStyle4Pokemon4SamMobile4SouthChinaMorningPost4Variety4WSB-TV4BellofLostSouls3DroidLife3GamesIndustry.biz3Hackaday3HollywoodReporter3HuffPost3Nature3PokeBeach3RockPaperShotgun3RPGSite3SeattleTimes3SFGATE3SlashGear3WhatHi-Fi?3404Media26abcPhiladelphia280Level2ABC7LosAngeles2AndroidCentral2AZFamily2Benzinga2Currently2DCRainmaker2DigitalFoundry2EventHubs2Futurism2GameInformer2HouseDigest2InsiderGaming2Jalopnik2LosAngelesTimes2Lifehacker2PCMag2qz2SimsCommunity2Slate2TimeExtension2TODAY2TweakTown2YahooFinanceUK2WindowsCentral224/7WallSt.1BusinessInsiderAfrica1Alternet1AndroidHeadlines1ArizonaSports1BikeRadar1BloodyDisgusting1BostonGlobe1BusinessTimes1Yahoo!FinanceCanada1CalMatters1CarBuzz1cbn1ChromeUnboxed1Chron1ClaimDepot1ColoradoSun1Skin.ClubCommunity1ChristianScienceMonitor1CyberSecurityNews1Cyclingnews1DailyDownforce1DaringFireball1DarkHorizons1Decrypt1Defense1denver71Designboom1DigitalCameraWorld1DirtonDirt1DSOGaming1Electrek1empireonline1GameGPU1erictopol.substack1Fangoria1FOX191DetroitFreePress1GameRant1GameWorldObserver1GAMINGbible1GamingOnLinux1AAAGasPrices1GearPatrol1GeekWire1GeekyGadgets1Global1Gothamist1Hackster.io1Hodinkee1HoustonChronicle1Independent1InterestingEngineering1Invezz1MacObserver1Magic:Gathering1MakeUseOf1Mashed1Mercury1MLive1MorningBrew1Motorsport1MP1st1MyNintendo1BloombergLaw1SemiAnalysis1Newsshooter1Newsweek1NintendoWire1nrn1CrudeOilPricesToday1OneMileataTime1OregonPublicBroadcasting1OregonLive1PageSix1PersonaCentral1politico.eu1QuantaMagazine1Realtor1Road&Track1RockstarINTEL1Salon1SeattleRed1Semafor1SanFranciscoChronicle1SimpleFlying1GhostHowls1SlippedDisc1SoraNews241SpaceNews1YahooTech1the5krunner1DailyBeast1DailyMeal1Drive1Hacker1Hindu1Intercept1Register1Times1TimesofIndia1TimesUnion1TMZ1TopGear1PCMagUK1UploadVR1VisualCapitalist1Vulture1WCVB1WFMZ1WHYY1WKYT1YGOrganization1YourTango1
  1. 001Hacker NewsSEP · 24English

    Jev deserves hype but not the type its getting

    A comprehensive benchmark compares Jev, TypeSafe AI's rubric-conditioned classification model, against Claude Haiku 4.5, Claude Sonnet 5, and OpenJev across ten test suites. Jev achieves 50-116x cost reduction and superior accuracy on core classification tasks, though advantages concentrate in specific architectural domains; Sonnet unexpectedly performs best on narrow tasks but worst on batched multi-question grading.

    By YidiDev
  2. 002Hacker NewsSEP · 24English

    iPhone 18 Pro Max SSD drops lower than hard drive at 1.1 MBs during heavy writes

    Apple's iPhone 18 Pro Max with 1TB QLC NAND storage shows significant performance drops under sustained heavy workloads, with write speeds falling to 1.1 MB/s when cache is exhausted and storage nears capacity. QLC's higher density enables larger capacities but sacrifices speed compared to TLC storage, though typical user tasks like browsing and photography remain unaffected.

    By Kunal Khullar
  3. 003Hacker NewsSEP · 23English

    Opus 5.5 Scores 75.6% on Part Catalog Bench

    Claude Opus 5.5 scored 75.6% on Part Catalog Bench, placing second behind GPT-6 Astra. Two other models, GPT-6 Sol and GPT-6 Luna, scored 55.5% and 21.0% respectively on the same 119-question benchmark.

    By thomas_ma
  4. 004Hacker NewsSEP · 23English

    New model gev outperforms jev and takes images as input too

    GEV model outperforms JEV in head-to-head benchmark comparisons and adds image input capability. The evaluation uses paired testing methodology to compare model performance across identical problems with controlled parameters.

    By ljlolel