Meta announces Llama 3.3 70B, a text-only model that Meta claims can deliver the performance of its largest Llama model at a lower cost
7.0M — 2,040 … The fine-tuning data includes publicly available … Markus Kasanmascheff / WinBuzzer : Meta Unveils New Llama 3.3 70B AI Model with Higher Cost-Efficiency Carl Franzen / VentureBeat : Meta launches open source Llama 3.3, shrinking powerful bigger model into smaller size Ananya Gairola / Benzinga : Meta Announces Llama 3.3 70B That Outperforms OpenAI's GPT-4o, Google's Gemini, and Amazon's AI Models Siddharth Jindal / Analytics India Magazine : Llama 3.3 Just Made Synthetic Data Generation Effortless Arjun Sha / Beebom : Meta's New Llama 3.3 70B Model Matches 405B's Performance at a Lower Cost Andrew Hutchinson / Social Media Today : Meta Launches New Llama AI Model, Building Towards the Next Stage Colin Kirkland / MediaPost : Meta Launches New, ‘Cost-Efficient’ Llama AI Model Maria Deutscher / SiliconANGLE : Meta releases efficiency-optimized Llama 3.3 70B large language model Corbin Davenport / How-To Geek : Your PC Can't Handle Meta's New Llama AI Model (Probably) Chris McKay / Maginative : Meta Wraps Up 2024 with the Release of Llama 3.3 Chris Ciaccia / Seeking Alpha : Meta unveils new Llama 3.3 AI model (NASDAQ:META) Mirza Samad : Meta's Llama 3.3: The Evolution of Open-Source Large Language Models John Quintet / iPhone in Canada Blog : Meta AI Nears 600 Million Active Users, Llama 3.3 is Here Karissa Bell / Engadget : Meta AI has ‘nearly’ 600 million monthly users Kalley Huang / The Information : Meta Releases New Version of Llama 3 Matthias Bastian / The Decoder : Meta's new Llama 3.3 aims for efficiency with lower computing needs Kylie Robison / The Verge : Meta launched a new ‘cost-efficient’ AI model.A new Llama 3.3 70B model dropped today, and Meta's VP of genAI said it matches the performance … Asif Razzaq / MarkTechPost : Meta AI Just Open-Sourced Llama 3.3: A New 70B Multilingual Large Language Model (LLM) Threads: Jonathan Ross / @jonathanrossgroq : (2/5) Today, our partner Meta released its latest version of Llama-3.3-70B-Instruct. And to all those who speculated that the industry had hit the wall - maybe some have, but Meta hasn't yet. 😉 This is a big deal. Casey Weinstein / @repweinstein : Are you using renewable energy to power the 2GW+ data center? Brian Penny / @thebrianpenny : What's considered a monthly active? Like once a month while browsing your site, I may see an AI overview. I may accidentally use your AI while trying to search. Are those activities included? If so, that 600m number is laughably low considering your userbases on Facebook and Instagram. Jonathan Ross / @jonathanrossgroq : (5/5) The new Llama-3.3-70B model launched and is now available to all 645,000 GroqCloud™ developers as of this morning. Go cook, and don't forget to share what you build here. Thank you for making GroqCloud™ the #1 API for fast inference! This is only just the beginning. Jonathan Ross / @jonathanrossgroq : (4/5) It's also significantly less expensive and faster than the larger model. Meta continues to push the lead in open weight innovations, and is keeping the pressure high for proprietary model providers to attempt to keep ahead of the giant wave of open. Casey Newton / @crumbler : Meta saying Llama has 600 million monthly actives feels similar to when it says Threads has 275 million monthly actives. In short: who and where are these people? Jonathan Ross / @jonathanrossgroq : (3/5) Though ~1/5th the size of Llama 3.1-405B, benchmarking shows Llama-3.3-70B performing neck and neck, and in many crucial cases substantially out performing the larger model (Instruction Following, Coding, Math, etc.), a suitable replacement for a majority of workloads. X: @aiatmeta : As we continue to explore new post-training techniques, today we're releasing Llama 3.3 — a new open source model that delivers leading performance and quality across text-based use cases such as synthetic data generation at a fraction of the inference cost. [image] Ahmad Al-Dahle / @ahmad_al_dahle : Introducing Llama 3.3 - a new 70B model that delivers the performance of our 405B model but is easier & more cost-efficient to run. By leveraging the latest advancements in post-training techniques including online preference optimization, this model improves core performance at [image] Awni Hannun / @awnihannun : Llama 3.3 70B 4-bit runs nicely on a 64GB M3 Max with in MLX LM (~10 toks/sec). Would be even faster on an M4 Max. Yesterday's server-only 405B is today's laptop 70B: [video] Rihard Jarc / @rihardjarc : @TidefallCapital Sure but even if the “real” number is half that it is still impressive. Also it is a way of reminding people that a search possibility is there. Soon it will form a habit and people will gradually use it more and more. If somebody knows how to introduce new features/format shifts Chris Paxton / @chris_j_paxton : Congrats on the launch, love seeing more open models But i really wish they had this Qwen comparison, numbers seem very similar (often better to stick with Qwen) [image] @arm : Congrats on Llama 3.3 70B, @AIatMeta! 👏 This 🆕 text-only model, available on Arm CPU, brings similar performance to the Llama 3.1 405B model 👉 so you get higher quality and performance at a fraction of the cost. @lmstudioai : 🥳 Llama 3.3 70B is available in LM Studio now! Matches the performance of Llama 3.2 405B 🚀 Search “Llama 3.3” in the app, or run: lms get llama-3.3 Poe / @poe_platform : Llama 3.3 70B is now available on Poe! It rivals the performance of the much larger Llama 3.1 405B, generates high-quality text responses, supports multiple languages, and has a 128k context window. (1/2) @metafordevs : Today we're releasing Llama 3.3 70B which delivers similar performance to Llama 3.1 405B allowing developers to achieve greater quality and performance on text-based applications at a lower price point. Download from Meta. [image] Matei Zaharia / @matei_zaharia : Congrats to the Meta team on Llama 3.3. Open weight models are advancing so rapidly and the cost to get this performance is quickly going down. We're thrilled to let users serve and customize this on @databricks starting next week. Kyle Corbitt / @corbtt : Meta just released Llama 3.3 70B—they claim benchmarks similar to Llama 3 405B, but in a model 20% the size. It's already available as a base model on OpenPipe, and we'll release benchmarks as a fine-tuning base model soon. 🫡 [image] Simon Willison / @simonw : “a new 70B model that delivers the performance of our 405B model” is exciting because I might just be able to run a quantized version of the 70B on my 64GB Mac - looking forward to some GGUFs of this @groqinc : Meta is challenging “death of scaling law” rumors with Llama 3.3 70B. They're defying traditional scaling limits, improving models without increasing parameters or changing the fundamental model architecture. Quality matters, not just quantity. https://hubs.la/... [image] @amd : Discover the enhanced Meta Llama 3.3 70B instruct model optimized for AMD compute, driving exceptional performance on text apps with advanced reasoning, math, general knowledge, and instruction following. Zach Horn / @zacharyhorn : Meta Llama 3.3 70B just released an hour ago. Coming to the Akash Chat web app and API today. Stay tuned. @reach_vb : BOOOOM! Meta released Llama 3.3 70B - 128K context, multilingual, enhanced tool calling, outperforms Llama 3.1 70B and comparable to Llama 405B 🔥 Comparable performance to 405B with 6x LESSER parameters ⚡ Llama 3.3 70B vs 405B: > GPQA Diamond (CoT): 50.5% vs 49.0% > Math [image] Yuchen Jin / @yuchenj_uw : We @hyperbolic_labs are now serving Llama 3.3 70B released by @AIatMeta today in BF16! ❤️🦙 > This 70B model delivers similar performance to Llama 3.1 405B —> cheaper and faster! > 128K context, multilingual. > It's up in @huggingface AnyChat, thanks to @_akhaliq and the [image] @kimmonismus : Ok, I didn't see that coming: Llama 3.3 70b has been released and is roughly equivalent to the performance of LLama 3.1 405b! That's a big surprise. It seems that OpenAI is so far ahead that others now have to follow suit. Good for us, more of the same. Anthropic, do you have [image] @togethercompute : 🦙 Big leap forward for Llama on Together AI! 🦙 We're rolling out support for the new Llama-3.3-70B-Instruct model! 🦙 This version delivers similar capabilities to the much larger Llama 3.1 405B model, but at a fraction of the cost. [image] Rowan Cheung / @rowancheung : AI NEWS: Meta just unexpectedly dropped Llama 3.3—a 70B model that's ~25x cheaper than GPT-4o. Plus, Google released a new Gemini model, OpenAI reinforcement finetuning, xAI's Grok is available for free, Copilot Vision, ElevenLabs GenFM, and more. Here's what you need to know: Paul Couvert / @itspaulai : Meta has just released Llama 3.3 70B which is more powerful than GPT-4o and 25x cheaper. Yes. 70B and better than GPT-4o. This model is also as powerful as the 405B version of Llama 3.1. Open source is really winning at every level. [image] Robert Scoble / @scobleizer : “this model improves core performance at a significantly lower cost” Moore's law just got one of those new @SpaceX rocket engines. Bill Gurley / @bgurley : More amazing open-source innovation! Would be both cheaper and faster. Florian Reifschneider / @flo_re2003 : Yes! @AIatMeta keeps delivering. I'd much rather see 12 days of Christmas releases from the open-source community than from @OpenAI to be honest Mikel / @mikelecheve : Meta's latest AI marvel, Llama 3.3, packs the punch of a 405B model into a 70B frame! 📦✨ With cutting-edge post-training optimization, this model isn't just accessible—it's a game-changer for open source enthusiasts looking to leverage top-tier AI capabilities without breaking Ben Sprecher / @bensprecher : The AI news cycle is now measured in RPM. Arjun Ram / @arjunram : Interesting 3 way battle in play and @finkd is playing all three. Model Size, Data distillation & GPU Size. Ryan Morrison / @ryanmorrisonjer : SOTA AI that can run on high-end consumer hardware. Prasanna Lahoti / @_prasannalahoti : meta is so back Vansh Kharidia / @vanshkharidia : Meta shipping the shit on Friday. I still believe 405B failed because of quantization issues and not because the model itself wasn't that good. LinkedIn: Benjamin Marie : Meta just released Llama 3.3 70B Instruct. The model seems significantly better than Llama 3.1 70B and even outperforms Llama 3.1 405B on some benchmarks! …