Archive · 20 Aug 2026
Enterprise AI’s Next Phase Focuses on Managing Model Evolution
Companies are using AI tools for many important jobs.
These tools can help people search, summarise, check and create information quickly.
But AI models keep changing as new versions are released.
A newer model may be more accurate or understand more information.
It may also cost more or behave differently.
This means companies cannot treat an AI upgrade like a simple software update.
They may need to retest instructions, safety checks and work processes.
The report says businesses should plan for these changes and track whether AI produces useful results.
Straive says enterprise AI is shifting from model deployment to managing continuous upgrades, costs and performance.
Organisations now use large language models for research, customer support, knowledge management, software engineering and enterprise search.
New model releases can improve reasoning and accuracy while changing pricing, output quality and model behaviour.
Model migrations may require companies to reassess prompts, evaluation data, safety measures, workflows and business rules.
The report recommends workflow-specific model selection, continuous budgeting, token-cost management and measuring returns through business outcomes.
- Who
- Straive and its Business Head for Science & Research, Digitalized Operations, Srinivasan Govindarajan.
- What
- A Straive report says the next phase of enterprise AI will focus on managing model evolution, costs and performance rather than simply deploying models.
- When
- August 20; the dateline identifies the report as released in New Delhi.
- Where
- New Delhi, India.
- Why
- Foundation models are continuously changing, and new versions can alter capabilities, pricing, output quality, behaviour and operating costs.
- Report author
- Straive
- Current enterprise AI uses
- Scientific research, customer support, knowledge management, software engineering and enterprise search
- Pricing model
- Foundation models generally use consumption-based pricing based on factors such as input and output tokens, reasoning tokens, search operations, tool calls, retrieval depth and context-window size.
- Example input pricing
- The report says Gemini 2.0 Flash costs USD 0.10 per million input tokens, compared with USD 0.30 for Gemini 2.5 Flash and USD 1.50 for Gemini 3.5 Flash.
- Example output pricing
- The report says output-token pricing rises from USD 0.40 per million tokens for Gemini 2.0 Flash to USD 2.50 for Gemini 2.5 Flash and USD 9.00 for Gemini 3.5 Flash.
- Reasoning settings
- Changing reasoning settings alone can create cost differences of up to four times, even when other application parameters remain unchanged.
- Recommended approach
- Treat model migration as a business process, budget for continuous evolution, manage token economics and measure returns through business outcomes.