![DeepSeek Market Share Changes](


alt="Article illustration 2"
loading="lazy">

) *DeepSeek's open-source dominance peaked at 80% but has since halved amid rising competition. (Source: OpenRouter State of AI Report 2025)* Despite open-source AI models offering staggering cost advantages—often **10-100x cheaper** than proprietary counterparts—new market analysis reveals established providers like OpenAI, Anthropic, and Google retain formidable pricing power. Data from OpenRouter's 2025 State of AI Report paints a picture of a bifurcated market where economic logic diverges sharply between enterprise and consumer applications.

The Pricing Paradox Explained

Open-source models have maintained a stable **22-25% market share** throughout 2025, briefly spiking to 35% during a surge of Chinese model adoption mid-year. This stability persists despite massive cost differentials, exposing weak price elasticity in core enterprise segments. As the report notes: *"Proprietary providers retain pricing power for mission-critical applications, while open ecosystems absorb volume from cost-sensitive users."*

Market Dynamics: The Rise of Chinese Models & Use Case Specialization

The open-source landscape has undergone dramatic shifts. DeepSeek, which commanded nearly **80% of OSS market share** in early 2025, has seen its dominance halved to around 40% as models like Qwen gain traction. Concurrently, clear product-market fit has emerged:

  • Programming dominates enterprise workflows: Coding accounts for 60% of Anthropic’s usage, 45% of xAI’s, and remains a top-2 use case for every major provider except DeepSeek.
  • Roleplaying fuels consumer growth: DeepSeek’s 80% roleplay usage illustrates how cost sensitivity shapes entertainment-focused adoption.
  • Scientific niche endures: OpenAI maintains a 20% science-focused user base—a legacy of early academic adoption.






















































Provider #1 Use Case % #2 Use Case %
Anthropic Programming 60% Roleplay 10%
xAI Programming 45% Technology 15%
Qwen Programming 27% Roleplay 18%
Google Roleplay 25% Programming 20%
OpenAI Programming 22% Science 20%
DeepSeek Roleplay 80% Programming 5%

Retention Reveals Workflow Stickiness

![Model Retention Comparison](


alt="Article illustration 3"
loading="lazy">

)
Claude 4 Sonnet and Gemini 2.5 Flash show significantly stronger retention than competitors. (Source: OpenRouter State of AI Report 2025)

Retention metrics underscore the specialization trend. While most models lose 60-70% of users within the first month, Anthropic’s Claude 4 Sonnet and Google’s Gemini 2.5 Flash achieve 40-50% Month 1 retention—far exceeding GPT-4o Mini and DeepSeek R1 (25-35%). This suggests models deeply embedded in professional workflows (like coding) develop stronger user loyalty than those used for transient entertainment.

The data ultimately reveals a market segmented by willingness-to-pay: Enterprises prioritize precision and reliability for critical tasks like software development, insulating premium providers from price competition. Meanwhile, open-source models capture volume-driven, cost-sensitive segments like roleplay—a divergence ensuring both ecosystems thrive but under fundamentally different economic rules.

Data Source: OpenRouter State of AI Report (2025). Analysis based on November 2025 usage data.
Original Report: Tom Tunguz - OpenRouter Insights