Open vs. closed · tracked daily
Open-source AI now handles 78% of real usage
As of September 22, 2026, open-weight models account for 78.3% of real token volume routed through OpenRouter, versus 20.3% for closed models - up 8.9 points since July 5, 2026. This is measured directly from production traffic, not a survey or an estimate, and is tracked daily rather than published as a one-off snapshot.
- Open-source share - 78.3%
- Closed share - 20.3%
- Best open model - Qwen: Qwen3.8 Max (53.4)
- Best closed model - Anthropic: Claude Opus 5.5 (batch) (57.6)
Usage share, open vs. closed
- Open-source
- Closed
Index date September 22, 2026Data through 00:00 UTC September 23, 2026Calculated 02:00 UTC
The quality gap
The best open model (Qwen: Qwen3.8 Max) currently scores 53.4on Artificial Analysis's Intelligence Index, versus 57.6 for the best closed model (Anthropic: Claude Opus 5.5 (batch)) - a gap of 4.2 points. This gap is tracked daily and will show whether open models are closing it over time.
- Best open
- Best closed
"Open" / "closed" is classified by publisher, not model-by-model license text - a curated author list (e.g. Meta, DeepSeek, Mistral, Qwen = open; OpenAI, Anthropic, Google Gemini, xAI = closed), documented in server/weights-class.ts. A small number of niche providers aren't yet classified and are excluded from the totals.
Frequently asked questions
Are open-source AI models catching up to closed models?
As of September 22, 2026, open-weight models handle 78% of real token volume on OpenRouter, up 8.9 percentage points since July 5, 2026. The best open model (Qwen: Qwen3.8 Max) trails the best closed model (Anthropic: Claude Opus 5.5 (batch)) by 4.2 intelligence-index points.
What share of AI usage is open source right now?
78.3% of real OpenRouter token volume went to open-weight models as of September 22, 2026; 20.3% went to closed models.
What is the best open-source AI model right now?
Qwen: Qwen3.8 Max currently leads open-weight models with an Artificial Analysis Intelligence Index of 53.4.
Source: OpenRouter (openrouter.ai/rankings), as of September 23, 2026.
Benchmark scores: Artificial Analysis (artificialanalysis.ai) via OpenRouter (openrouter.ai/rankings).
Methodology: www.smophy.ai/benchmark/methodology
Token counts originate from each provider's own tokenizer and are not directly comparable across providers.
