|
The next phase of enterprise artificial intelligence (AI) will be less about deploying the latest models and more about managing their continuous upgrades, costs and performance, according to a report by Straive.
The report said generative AI has moved beyond experimentation, with organisations using large language models (LLMs) across areas such as scientific research, customer support, knowledge management, software engineering and enterprise search. It said, "The next phase of enterprise AI is therefore no longer about deploying models. It is about managing their evolution". According to the report, AI has helped teams complete tasks such as searching, summarising, validating and creating content in minutes instead of hours. However, as companies expand their use of AI, they are facing a new challenge because foundation models are continuously changing. The report said every new model release can bring improvements in reasoning, context windows, multimodal capabilities and accuracy, but can also change pricing, output quality and how the model behaves. It said moving from one model generation to another is not similar to a traditional software update. Companies may have to reassess prompts, evaluation datasets, validation systems, safety measures, business rules, workflows and operational performance. The report also highlighted the changing cost structure of AI models. Unlike traditional software licensing, foundation models generally operate on consumption-based pricing, where costs can depend on input and output tokens, reasoning or "thinking" tokens, search operations, tool calls, retrieval depth and context window size. For example, the report said moving from Gemini 2.0 Flash to Gemini 2.5 Flash increases input token pricing from USD 0.10 to USD 0.30 per million tokens, while output token pricing rises from USD 0.40 to USD 2.50. Gemini 3.5 Flash increases these costs further to USD 1.50 for input tokens and USD 9.00 for output tokens. Srinivasan Govindarajan, Business Head, Science & Research, Digitalized Operations at Straive said, "Unlike conventional software licensing, foundation models operate on a consumption-based pricing model where every interaction generates costs. Those costs are influenced by multiple billing drivers, including input tokens, output tokens, reasoning ("thinking") tokens, search operations, tool calls, retrieval depth, and context window size". The report also said changing reasoning settings alone can create cost differences of up to four times, even when other application parameters remain unchanged. It therefore urged companies to select models based on the requirements of each business workflow instead of simply choosing the newest or cheapest model. The report recommended treating model migration as a business process, budgeting for continuous AI evolution, managing token economics, designing AI systems for cost efficiency and measuring returns through business outcomes. "The future of enterprise AI will be defined not by the intelligence organizations can access, but by how efficiently they can convert that intelligence into measurable business value," the report outlined. (ANI)
|
