BotBeat
...
← Back

> ▌

MetaMeta
INDUSTRY REPORTMeta2026-07-03

Open Source LLMs Now Account for One-Third of All Token Volume, Report Finds

Key Takeaways

  • ▸Open source LLMs reached approximately 33% of total token volume on major platforms by late 2025, proving viability for production workloads
  • ▸Chinese open source models surged from 1.2% to as high as 30% of weekly token volume, signaling major regional shifts in AI development
  • ▸DeepSeek dominates open source LLMs with 2.6x more tokens than its nearest competitor, but usage has spread across 5-7 competitive models, indicating a healthy ecosystem
Source:
Hacker Newshttps://openllmstack.com/blog/open-source-llm-statistics/↗

Summary

Open source large language models have matured rapidly from experimental technology to the backbone of production systems, according to a comprehensive statistical analysis. By late 2025, open source models accounted for approximately one-third of all token volume on OpenRouter, a major unified API platform, based on analysis of over 100 trillion tokens spanning November 2024 through November 2025.

The ecosystem shows rapid diversification and competitive spread. DeepSeek emerged as the dominant open source model family with 2.6 times more tokens than Qwen, the second-largest open source competitor. However, usage has become increasingly fragmented, with no single model commanding more than 20-25% of open source token volume by late 2025—a sign of a maturing, competitive ecosystem. Chinese open source models surged dramatically from as little as 1.2% of weekly token volume in late 2024 to nearly 30% in some weeks, though they averaged 13% over the full study period.

Developer adoption of AI tools has become nearly universal, with 84% of professional developers using or planning to use AI tools according to the 2025 Stack Overflow Developer Survey, up from 76% the previous year. OpenRouter itself grew from 2.5 million to over 8 million developers in roughly one year and reached an annualized run rate of 1.5 quadrillion tokens by May 2026, indicating explosive growth in the sector.

  • Developer adoption of AI tools reached 84% among professional developers, with 51% using AI tools daily
  • OpenRouter scaled from 2.5 million to 8+ million developers in one year, reaching 1.5 quadrillion tokens annualized by May 2026

Editorial Opinion

The data shows that open source LLMs have crossed a critical threshold from niche innovation to mainstream infrastructure—this is no longer about experimentation, it's about production systems at scale. The fact that no single model dominates the market is actually healthy; it signals a competitive ecosystem where developers have genuine choice and can select models optimized for their specific use cases. The surge of Chinese models demonstrates that open source is truly global, and the ability to rapidly iterate and deploy is increasingly decoupled from capital advantages. Proprietary model providers can no longer rely on a closed-model moat; the competitive pressure is real and intensifying.

Large Language Models (LLMs)Generative AIMarket TrendsOpen Source

More from Meta

MetaMeta
UPDATE

Zuckerberg Signals Slower-Than-Expected Progress on AI Agent Development

2026-07-02
MetaMeta
INDUSTRY REPORT

Meta's Cloud Push Overshadows Bigger Story: Saudi Arabia's Data Center Dominance

2026-07-02
MetaMeta
UPDATE

WhatsApp Username Rollout Faces Impersonation Scrutiny, Regulatory Warning in India

2026-07-02

Comments

Suggested

IntelIntel
FUNDING & BUSINESS

Yann LeCun's AMI Labs Raises $1 Billion to Develop Post-LLM AI Architecture

2026-07-03
AnthropicAnthropic
UPDATE

Anthropic Introduces Advanced Analytics and Cost Controls for Claude Enterprise

2026-07-03
Multiple AI CompaniesMultiple AI Companies
INDUSTRY REPORT

What Is Agentic AI Today, and What Do We Want It to Be?

2026-07-03
← Back to news
© 2026 BotBeat
AboutPrivacy PolicyTerms of ServiceContact Us