AI Startups Shift Pricing Models for Enterprise
AI startups are rethinking their pricing strategies, moving away from flat seat-based models toward usage-based pricing for tokens or API calls. This shift is driven by the variable costs of model inference. As startups transition from product-led growth to high-touch enterprise sales, they are also bundling RAG, vector databases, and LLM serving into unified contracts to simplify procurement.
- Enterprise search competitor Glean avoids usage-based pricing, opting for a per-user, per-month model that starts around $45-$50 per user. They also require minimum annual contracts of $50,000-$60,000, often mandating at least 100 users. - Competitor Hebbia, which targets complex financial and legal research, also uses a per-user licensing model, with reports suggesting costs of $15,000-$20,000 per user annually. This model is designed for deep research workflows where Hebbia claims it can automate up to 90% of the work. - Foundation model provider Cohere prices its Command R+ and Command A models at $2.50 per million input tokens and $10.00 per million output tokens, designed for high-performance and agentic tasks. In contrast, their more balanced Command R model is significantly cheaper at $0.15 per million input and $0.60 per million output tokens. - OpenAI’s enterprise offerings provide a choice between pay-as-you-go API usage and per-seat licensing for ChatGPT Enterprise. The per-seat model offers cost predictability for large teams, while the API model is better suited for variable workloads and experimentation. - Anthropic recently shifted its enterprise pricing, moving away from high per-user fees to lower seat costs combined with mandatory consumption commitments. Their "Claude Code" plan for technical users is priced at $20 per user per month, while a "Claude.ai" plan for business users is $10 per user per month. - The cost of inference, a key driver of usage-based pricing, is seeing rapid decreases, with some analyses showing a 50x per year price decline between 2020 and early 2025 for equivalent performance. This trend is fueled by smaller, more efficient models like Google's Gemini 2.5 Flash and Anthropic's Claude 3.5 Haiku. - The adoption of AI in enterprise procurement is accelerating, with 94% of procurement executives using generative AI at least weekly. However, only 4% of procurement teams have achieved large-scale deployment, indicating a significant growth area for AI startups. - Selling AI to enterprises often requires an "educative sale" to bridge the gap between the technology's complexity and the buyer's business needs. Common enterprise concerns that need to be addressed during the sales process include data privacy, security, and the complexity of implementation.
