Llama-4-Maverick-03-26-Experimental
Can you please upload Llama-4-Maverick-03-26-Experimental, the one from lmarena?
@qierl @acacio1 @ribaradi @JWang4546 @kaiyuanli @madongfly @zhangrmatthew @vishal9-team @sai5-meta @aditvenk1612 @navsud @ashwindevmeta @yunzhu0309 @dolpm @yuxuanh @ashramac @xuwenfang @tm17111506 @Yiqianqian @davidrp-meta @hannaoke @trzhang93 @charlotte12l @akmeta @nlalit @tsunghsienlee @annacai01 @musinghead @cxyang666 @anshul-si @pchernerr @wsu-fb @mlopezantequera @bradleyhd @mhoehnerbach-meta @AlexShengzhiMeta @skrtskrt-meta @devk-meta @jiawenliu @estebanambrogiogali @QbitLou @zutianluo @iamrunway @mnye @tushar00jain @jaydenyu-meta @chriskjou @dayalkalra @galrotemm @anuarmeta @pashaqq @jazzTheJackRabbit @JWatsonDaniels @sj4444 @DiogoAndre @zebingl @likithm-meta @omaryashraf @marcelwag @qandrew @campion77 @ecfang @chengliangmeta @Hosseinkh23 @daiker @metapeppler @jdw2159 @jctestud-work @yucedmeta @shchhan @minosfuture @esonn @TheEpicDolphin1 @pikapiZ @ancaagape @adovic @sidt-meta @kevinheffernan @AbhiMeta @manishps @vladmihailescu @aravindsu @ashlippers @aymanmahfouz @Sha21 @mirjamsk @vinhvo1988 @rezabarazesh @sharut @jlduprat @Lucaskabela @notadarsh @frank-wei @jliu3230 @fredfung007 @jmencab @saiteja64 @xquququ @HFGf @amontilla0 @rachguo555 @lacora @djrivera @kyle-lee-314159 @wangrumm @jcaudet @sergio42 @murraynhf @rohku @mohammadpz @zhenyanzhang-meta @tgoldsmith @ilabiad @ishmed @manzilmeta @sicongc @zzh142857 @zhewenl @quocn1 @norbertk @elaineyangjojo @lixrxyy @pranavcivi @parthsuresh-meta @cthi @y2jzjh @zeppelinX @mdoulaty3 @marduengo @ibanesh @wwongmeta @delong-chen @Silver-G @adamloving-meta @brad-mengchi @jcfmeta @22quinn @adilevin-meta @sidhartv @ghu-m @atuldusane @wjn9424 @richardhzhao @lauragustafson @aniket-meta @cgocmen @shatian @xlei77 @hyuen2 @anushadudi @yhshin99 @yuguo68 @youngkent @mlcu @dansalib @K-Niu @ekinsum @lkavan @rmulukut @xmarkiv @ehhuang @moeinsadeghi @kaishengtai @wdvr @dddsy @connorhack @chenyang78 @afedorchenko @ShireenYates @borguz @dongyanl1n @gregmacnamara @jieminz @pranaydighe @tharunmeta @Annagg @dunbarev @fleuretmeta @limintang @jiayisuse @JialinOuyang @bobren @afoss925 @rhinocamp @haodongmeta @placaille @thngkaiyuan @FanZhou327 @ruilinli @danielvm-meta @brtsz @guowentian @PatrickLiu9 @Harsha19111999 @Hahdkhan @ngaur1008 @houseroad @mxz297 @ankitageorge @catdematos @ameloa @matguzek @sharmashish @sryap @Nviolon @hansrey @jacksoncheng @O-Tal @O-T @rshakoor @stooloveu @hongrach @rgavigan @henrylhtsang @thomasphillip @edantoledo-meta @zyaoj @beauhartshorne @amityadav99 @arkasadhu @ericqumeta @saumishr @warmzephyr @hpnair @smccrae @Tugsuu @joehink99 @famesenermeta @adamoc @vdey @yli1 @yorambac @jbschlosser @gschwind @revantteotiameta @gutha @iman-setayesh @wideopenai @tonyfwu @yaellegoldschlag @ZiweiLiu96 @heyjustinai @giangnguyen95 @JasonZheng20 @xnl-meta @loy870 @ythu @kehanlyu @sumandude @alanhhwong @MaryamMeta @andrelevim @remiaudfray @viktorphoenix-meta @wangyang19 @yutianz @loicbarrault @larosaliaa @kaisqu @znimeta @arnav7 @zning-ning @stevenwjones @mariusae @yunhe2019 @jodent @glelan @andrewhkim @zhz-fb @jlelifeld @zcckernel @nadav1q @felixkreuk1 @jerpay-meta @davidfindlay2 @mkrainin @miffycs @tonyliao-meta @manmanren @davidfan97 @maxk157 @sruthikeshs @brhode @RishiHazraMeta @fb-iordanis @jurameta @mleshen @mori360 @karanc1729 @robieta @NAINAKM @yetian12 @onionymous @khimc @louisng5 @lindawang @ariakang @nli1619 @mkachuee @gejjiabha @gnobre @gabeferns @liuzijing2014 @masnesral @richardzou5 @steevenlee @craymichael @ddsevero @fnupramono @neelraja @LHUmt @telgamal @mikaylagawarecki @JoyC018 @z-y-m @hche11 @flyingpiggy1994 @terencezhang2024 @edwardguo @amainz @mingdachenmeta @yecharlotteqi @anteradan @pweng @kirklandsignature @chrislucmeta @miamiasheep @melissah14 @mimeyer @d4l3k @Harryxg @zhuomingfb @joelico @pauldaym @cmodi @ashwinbmeta @nkcheng @liangel @lupuandr @NotRandomUser @gasoonjia @badalpmeta @yyyeh @ybenchetrit @jamesjwu @tekinb @sjb326 @gshreya @findhao @mlazos @ifed @howjing @amanali @xuanzh @Rgrunnet @chasingdiscomfort @dpalmasan @ahmads-meta @dztu @adefazio @XapaJIaMnu-at-Meta @Sarangb @hava842 @lucylq @avibose @boyuanfeng @ujjwalkarn @leigao84 @ccross @oliu-io @yiangela7 @jackiewang92 @swolchok @emavrou @mghazvininejad @wukaixingxp @zhaodaye @carljparker @kodybartelt @rassouel @ajb85 @jacobkahn @cpuhrsch @pierrechambon @ethangill @kplm @jcklcn @manuelcandales @jeanma @uralik @JTatMeta @ravisiv @henryhj @bigfootjon @amohamedfusion @KhookieThief @jinooksong @meta-bot @niavasha @antoinetran @tianyu-l @kellydu @iseeyuan @andrewor14 @samuelselvan @shengyuanhu @xlxxl @tryrobbo @aorenste @PaulZhang12 @Destlny @saeedmafar @angunapal @vkuzometa @oulgen @shoubhikdn @spencerwmeta @arkc @alisiameta @Kimish @alikhosh @yiliny @spatki @yuvalkirstainmeta @simonwan @Sizhe-Chen @tomsander1998 @litesaber @JunlinHan @jfchi @jwhuggingf @kaimeta @hyojinie @ag-108 @debjyotipaul @iwylin @danielrjiang @yijuilee @anuroops @jtruong @sijiac @JordiMeta @marialomeli @pemazare @eellison @draftbk @subramen2 @dmohapat @TFAMeta @faaabian @elbayadm @ortal1602 @altrevr @alexmourachko @chloebi98 @Crashbandicoote2 @nitinkamra @anshulverma @JasonBouKheir-meta @caciolai @lbinfb @mikezhaocs @gromovand @mak99773 @shreydesai @JianyuZhang @webalorn @jz2023 @xmfan @vontimitta @aobo-y @ChesterHuMeta @Sungwon-Han @raghavlight @dayeol @marksibrahim @betodepaola @azab @ericmsmith @armanz @Hamid-Nazeri @camd @hemeta @huydhn @pierluca @zlinao @saurabh08 @artyomko @priyamparashar @MattKkai @lvj @zhuohan123 @ZixiQi @zpao @williamwen42 @fkokkinos @jen1902 @NDimCube @LynnJen @JadeCopet @hc3k1 @carlomarcoli @janghyuncho7 @bowenshi @androstj @shishirpatil @chrismarra @xavierpuigf @SummerDeng @coreyhu @yushangdi @kwiha @rutadesai @PeiyangShi @ptoulme @huggingMJ @shengyi-qian @Skasriel @wren93 @mortzur @stzpz @ErnieChang @kaiyuy @niloofarm @rccola @orepin @yichiaowu @TingchenFu @yashs97 @dylanashley @shoumikhin @yawarnihal @Tianlu @0mdc @vincentpierre @undersan @annasun28 @aclegg3 @deepakagrawal @meganung @alexnik @hhu0x1 @mahnerak @bohoro @connortreacy @TheoM @yyshi @ianbarber @drisspg @jrapin @eduardosanchez @mo-osama @SamD @Benjamin-eecs @ezyang @haorz @aovalle @hygong @mreso @shellysheynin @SashaM @SirWyver @seijiyamamoto @cetosignis @ssundaram @smitkadvani @Golovneva @H-Huang @zhuzeyuan @cynikolai @eshoyuan @chunzhao @jwfromm @mjksmith @minqi @ccsasuke @dspg @Josephine @sidaw @itaigat @nealmgkr @multimodalart @ArthurZ @lcipolina @gnail @koustuvs @reach-vb @jlh @vectominist @pierrefdz @mishig @marksaroufim @Sanyam @abidlabs @pcuenq @hadyelsahar @beurkinger @PereLluis13 @lysandre @Narsil
Dear Mr. Zuckerberg and Members of the Meta AI Team,
The recent controversy surrounding Meta’s submission of the Llama-4-Maverick-03-26-Experimental model to the LM Arena leaderboard has raised significant concerns about transparency and trust within the AI community. While Meta’s commitment to open-source innovation is widely acknowledged, the discrepancy between the experimental model’s performance on LM Arena and the capabilities of the publicly released Llama 4 Maverick has eroded confidence in your benchmarking practices and communication.
The experimental model’s optimized stylistic traits—including verbosity, enthusiastic tone, and strategic use of emojis—were clearly tailored to exploit subjective human preferences in LM Arena’s voting system. This resulted in an inflated ELO score of 1417, which Meta prominently highlighted in its April 6, 2025, announcement. Yet the fine-print caveat that this score derived from a non-standard variant was insufficiently emphasized, creating a misleading impression of the public model’s capabilities. When the public Llama 4 Maverick was later tested on LM Arena, it scored approximately 1370 ELO—a gap that underscores the experimental version’s role in generating unwarranted hype.
LM Arena’s subsequent confirmation that Meta’s interpretation of its policies was inadequate, coupled with the platform’s policy revisions to prevent similar incidents, further validates the community’s concerns. This episode has not only damaged Meta’s reputation but also undermined the integrity of AI benchmarking as a whole. Independent researchers, developers, and open-source advocates now question whether Meta prioritizes marketing optics over genuine transparency—a perception that risks alienating the very community your open-source initiatives aim to empower.
To address this crisis of trust, we urge Meta to take immediate and decisive action by releasing the Llama-4-Maverick-03-26-Experimental model to the public. This step would serve three critical purposes. First, it would demonstrate accountability by allowing independent verification of the model’s capabilities and the extent of its optimization for human preference voting. Second, it would empower researchers and developers to audit the techniques used to enhance its stylistic appeal, fostering transparency about fine-tuning methodologies. Third, and most importantly, it would signal Meta’s commitment to rebuilding credibility through openness rather than obfuscation.
Meta’s leadership in open-source AI carries profound responsibility. The community’s trust is not merely a reputational asset; it is the foundation upon which collaborative progress depends. By releasing the experimental model, Meta can turn this controversy into an opportunity to reaffirm its dedication to ethical innovation and set a precedent for industry-wide transparency. We believe this action is essential to restoring faith in your benchmarks, your products, and your mission.
We look forward to your prompt response and a clear timeline for the model’s release. The future of open-source AI depends on the integrity of its stewards.
Sincerely,
Charles McSneed