AI BriefingMozilla Releases The State of Open Source AI 2026 Report
Key point
The performance gap for open models is narrowing and inference costs are plummeting, shifting the center of competition to the agentic layer.
Details
According to the 'The State of Open Source AI 2026' report released by the Mozilla Foundation, open weights models have nearly closed the performance gap with closed models, and the price of intelligence is dropping sharply.
Key Metrics and Current Status:
- Narrowing Performance Gap: Based on Chatbot Arena, the performance gap between top closed models and open models narrowed from 8.04% in 2024 to 0.5% in February 2025, and as of March 2026 it stands at 3.3%.
- Collapse in Inference Costs: The price per million tokens for GPT-4-class performance models has dropped 50-fold ($20 $\rightarrow$ $0.40) in 36 months. The pace of the decline accelerated notably with the emergence of Llama 3.1 and DeepSeek.
- Market Shift: Based on OpenRouter data, the share of tokens processed by open models has risen to about 1/3, indicating a substantial shift in actual traffic.
Strategic Shift: As performance differences among models themselves shrink, the core of competition is now shifting away from the models themselves toward the Agentic Harness layer that uses models to perform complex tasks.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Hugging Face CEO details lessons from first public agent cyberattack, advocates for open-source defense
NASA-IBM Releases Open-Source Multimodal Geospatial AI Model 'Lunar Foundation' for Lunar ExplorationTerence Tao Introduces SAIR 'Open Math Model' Initiative, Calls for Participation in Building Open-Source Math AI ModelsUAE Falcons AI's NSFW Classification Model Enters Top Ranks on Hugging Face with 50.8 Million Downloads in 28 Days