Mozilla CTO Says Open AI Trails Proprietary Models by Just 3%
Updated
Updated · Rest of World · Aug 13
Mozilla CTO Says Open AI Trails Proprietary Models by Just 3%
1 articles · Updated · Rest of World · Aug 13
Summary
Mozilla argues open AI is now close to the frontier: its July report said top open-source models lag proprietary systems such as ChatGPT and Claude by only 3%.
Raffi Krikorian said that narrowing gap is driving business adoption because companies want cheaper tools they can self-host, especially in regulated industries that need tighter control over data and infrastructure.
Mozilla also pointed to commercial momentum, saying Alibaba’s Qwen was downloaded more in February 2026 than the next eight models combined, even as consumer attention stays focused on OpenAI and Anthropic.
Krikorian said policymakers should treat AI as infrastructure rather than a rentable product, warning that backing a few likely IPO candidates could leave demand for open, customizable systems to others.
He acknowledged most leading alternatives are still open-weight rather than fully open-source, but said they remain the main path for local languages, community-specific tuning and broader participation outside the West.
With open AI capturing massive usage but minimal revenue, who will ultimately monopolize the billion-dollar agentic harness goldmine?
As Chinese models like Qwen dominate downloads, can Western enterprises truly secure sensitive data while relying on foreign open-weight infrastructure?