Should the most capable frontier models be released open-weight, or does safety require withholding at the bleeding edge?

Alexandr Wang · Jun 12, 2024

The most advanced frontier AI systems should be kept closed for geopolitical and military reasons, while less advanced models can and should remain open to generate economic value

Genuinely powerful systems are military assets, but models below that threshold — like Llama 3 today — are not yet military assets and mainly create economic value

Scope: the hard part is identifying where the line is and when we approach it; Llama 3 is currently below the line

37:42 20VC: Scale's Alex Wang on Why Data Not Compute is the Bottleneck to Foundation Model Performance, Why AI is the Greatest Military Asset Ever, Is China Really Two Years Behind the US in AI and Why the CCPs Industrial Approach is Better than Anyone Else's

Alex Schultz · Sep 5, 2025

Meta favors open source generally but at the absolute bleeding edge of capability open release may not be the right answer and would need reevaluating on safety grounds

Meta couldn't exist without MySQL, PHP, Memcached and Open Compute, but it's important to be sensible about safety at extreme capability levels

Scope: applies only to the most extreme frontier capabilities

32:56 20Growth: Meta CMO Alex Schultz on How All Founders Have to Change Their Marketing Playbook in a World of AI | Is AI Plateauing, What it Means if China Wins the AI Race and Why Zuck is a Generational Leader

Jonathan Ross · Sep 29, 2025

Anthropic should open source its previous-generation models to displace Chinese models

Prompt compatibility acts like software compatibility — users on an open Anthropic model can reuse prompts when they upgrade to the premium model, and infrastructure providers will drive costs down; this is exactly why GPT-OSS won adoption over Chinese models

30:45 20VC: OpenAI and Anthropic Will Build Their Own Chips | NVIDIA Will Be Worth $10TRN | How to Solve the Energy Required for AI... Nuclear | Why China is Behind the US in the Race for AGI with Jonathan Ross, Groq Founder

Chris Dixon · Mar 27, 2024

A dynamic internet and software economy is at serious risk of being lost, with the two central threats being efforts to ban open source AI and to ban blockchains

A dynamic internet economy is good for innovation and for the world, and he sees himself as working on behalf of little tech rather than big tech

46:22 20VC: a16z's Chris Dixon on Who Will Win the Next Generation of Venture, The Two Ways to Make Great Venture Investments and Find the Best Entrepreneurs & Why AI Will Strengthen the Position of the Incumbents Moving Forward

Chris Dixon · Mar 27, 2024

Open sourcing frontier AI models should be legal and each project should be free to choose its own strategy, rather than open sourcing being mandated or banned

Mistral, Meta's Llama and others should be allowed to open their code and weights; the current panic is driving registration requirements and export controls under the Biden executive order

Scope: not saying OpenAI specifically should be open sourced

47:38 20VC: a16z's Chris Dixon on Who Will Win the Next Generation of Venture, The Two Ways to Make Great Venture Investments and Find the Best Entrepreneurs & Why AI Will Strengthen the Position of the Incumbents Moving Forward

Martin Casado · Jul 28, 2025

The recent consensus treating open, transparent innovation as antithetical to safety was bizarro-land, and it's crazy for VCs to argue against open source

Pro-innovation sectors of the economy — VCs, Founders Fund, academia — had all somehow adopted a position against open transparent innovation

24:52 20VC: a16z's Martin Casado on Anthropic vs OpenAI: Where Value Accrues | Cursor vs Replit vs Lovable: Who Wins and Who Loses | The One Sin in AI Investing | Why Open Source is a National Security Risk with China

Sarah Tavel · May 6, 2024

Frontier AI models will be closed source and the frontier will be controlled by an oligopoly of a few players

Progress is currently compute constrained and each successive model is dramatically more expensive to train, so only a handful of heavily capitalized players can stay on the frontier

Scope: applies specifically to models 'on the frontier'

0:00 20VC: Benchmark's Sarah Tavel on Are Foundation Models Commoditising | Why Frontier Models Will Be Closed Source | Why the Value is in the Application Layer | The Future of AI is "Selling the Work" Not the Tools

Sarah Tavel · May 6, 2024

The frontier model landscape will end up an oligopoly rather than commoditizing

Progress is compute constrained and each successive model is more and more expensive to train, so only a few players can fund the frontier

32:48 20VC: Benchmark's Sarah Tavel on Are Foundation Models Commoditising | Why Frontier Models Will Be Closed Source | Why the Value is in the Application Layer | The Future of AI is "Selling the Work" Not the Tools

Sarah Tavel · May 6, 2024 · hedged

Models at the frontier will be closed source

Each step function costs roughly 10x more to train, and open sourcing requires believing you can capture value from that investment, which is very TBD

Scope: view shifts month to month; Meta making the huge training investment and open sourcing anyway could fundamentally change the game; some use cases don't need frontier capability, where open source is powerful enough and its flexibility compensates

34:30 20VC: Benchmark's Sarah Tavel on Are Foundation Models Commoditising | Why Frontier Models Will Be Closed Source | Why the Value is in the Application Layer | The Future of AI is "Selling the Work" Not the Tools

Clem Delangue · May 12, 2023

Elon Musk is right about the necessity of openness and transparency in AI, and it is valuable that he is putting that part of the conversation in the spotlight.

Openness and transparency matter not only for AI but for society.

Scope: Musk is a character who will say one thing and its opposite on the same day

19:22 20VC: Why The Future of AI Is Open Not Closed, Why We Are Years Away From AI Being Autonomous, Why AI Founders Do Not Need to Move to the Valley & Why Founders Should Not Meet Investors in Between Rounds with Clem Delangue @ Hugging Face

Micha Kaufman · Jun 9, 2025 · hedged

Because AI is such a fundamental technology engine, he'd rather it be open than closed

Its fundamental, engine-like nature argues for openness

Scope: 'would probably rather'

55:26 20VC: Fiverr CEO: 'If You're Not Adapting to AI, F* You. You're Done | Why "Time to Copy" is the Most Important Metric in Startups Today | Why 99% of AI Companies Today Will Die | Why Governments Will Take Control of AI with Micha Kauffman

Arthur Mensch · Apr 29, 2024

Keeping some models closed is justified as a commercial asset to grow the business and to cement strategic relationships with cloud providers, while Mistral still intends to lead in open source

The closed models can be licensed and sold, which grows revenue and gives cloud partners a reason to deepen the relationship, and unique assets plus a developer platform are what differentiate the company

25:29 Keeping some models closed is justified as a commercial asset and partnership lever while still leading open source

David Cahn · Aug 5, 2024

The open vs closed model safety debate is downstream of AGI timelines, and since AGI is not imminent there is little to worry about

If you think AGI happens tomorrow you're afraid; he thinks AGI might arrive when he and Harry are 90, and AI will mostly improve lives and productivity

38:46 Safety urgency is downstream of agi timelines so non imminent agi means little to worry about

Martin Casado · Jul 28, 2025

'Open source' is a misnomer in AI: the standard practice is to release the smaller model and keep the more capable one closed, which buys distribution and brand recognition without eroding the business

Unlike software, releasing a model doesn't let anyone replicate it—you'd have to recreate the data pipeline and the training pipeline—and labs are understandably reluctant to give away hundreds of millions or billions of dollars of training

33:38 Release only the smaller model for brand while keeping frontier capability closed

Cristóbal Valenzuela · Aug 28, 2023

Open is generally better, but open vs closed source is only one part of the stack and you can't build a company on that choice alone — the right answer depends on the product and value proposition

Open source gets you to a broader audience faster and brings outside perspectives because people build on top of the models, but different companies' value props will pull them in different directions

19:57 Open vs closed is only one part of the stack the right mix depends on product and value proposition

Harry Stebbings · Apr 4, 2025

OpenAI should open source immediately or it will die

50:21 Openai must open source immediately or risk irrelevance