Good morning leaders, today we dive into open source models and why they might be the next option for companies looking to spend smartly or not go over their tokens.
Hint maybe this is the meta AI playbook…
The Exec Summary
Open source AI models are no longer just a niche curiosity for developers; they have become a strategic imperative for leaders facing spiralling token costs and the need for greater architectural flexibility.
While “Big AI” like Gemini and ChatGPT often dominate the headlines, open source alternatives like Meta’s Llama and the increasingly powerful models from China, such as DeepSeek and Qwen, are offering a viable, high-performance path forward. As venture capital funding eventually cools and premium tier pricing rises, the ability to “own” your model through open source will be the difference between a sustainable AI strategy and a budgetary nightmare.
After recording we have seen a wave of AI startups move away from major models over to open source, including automation company Lindy
The Key 6 Points Discussed
Defining Open Source Accessibility: Unlike premium, “closed” software, open source models are coded and edited to be accessible and open, often funded by large organisations to provide a genuine alternative to proprietary systems.
The Meta Strategy: Meta is positioning itself to win the long game by open sourcing Llama, potentially shifting the entire user base toward their ecosystem as people seek high-quality output without the premium price tag.
The Token Budget Crisis: Many organisations are burning through annual token budgets in months. Open source provides a way to scale without the unpredictable and often uncontrollable costs of proprietary APIs.
The Performance Parity: Models like DeepSeek version 4 have reached a level of power where they can handle the majority of everyday business tasks perfectly well compared to their paid counterparts.
Navigating Governance and Bias: Open source models, particularly those from specific regions like China, may carry different biases or data transparency issues. It is vital to apply the same rigorous governance and testing used for any other enterprise tool.
The Rise of the “Wrapper” Economy: Many startups are now using open source models as a base and building proprietary “wrappers” or custom edits on top, a trend tech teams are likely to propose as a standard moving forward.
The AI Tools Mentioned
Meta Llama: Meta’s flagship open source large language model.
DeepSeek: A powerful Chinese model noted for its performance in version 4.
Qwen: Another prominent open source model within the industry.
Hugging Face: The platform and community hub for discovering and testing open source models.
Claude / ChatGPT / Gemini: The “closed” proprietary benchmarks used for comparison. There are localised versions (which run on machines and modern smartphones) of these models becoming available.
Firefox: Cited as a historical example of how the open source movement operates. They are now big advocates of local and smaller models
Here is a good podcast discussing AI, open source and the future of AI/ML
Shareable Insights
On Budgeting: “We need to evaluate open source models now. As VC funding slows, proprietary token costs will likely skyrocket, and having an open source alternative is our best hedge against price hikes.”
On Performance: “Don’t assume ‘free’ means ‘inferior.’ Modern open weight models are now powerful enough for most of our workflow needs. We should be benchmarking DeepSeek and Llama against our current paid stack.”
On the ‘Two Vs’: “Regardless of which model we use, our policy must be ‘Validate and Verify.’ We care about the output quality and accuracy, not just the name of the engine behind it.”
The Core Takeaway for Leaders
The future of AI is “horses for courses.” While proprietary models offer ease of use, open source provides the flexibility, cost control, and data sovereignty required for long term scaling. Leaders must move beyond the brand names and start testing open source alternatives to ensure their AI strategy is built on solid, sustainable ground.
Thanks and have a great day!
Danny & Jonathan
