The context
After an industrialisation phase of AI generated content projects (text, meta descriptions, SEO, GEO, translations…), without real measurement of return on investment, the priority for 2026 is different. With retailers launching more complex AI projects involving video content generation (Decathlon), image search (Sézane), etc., “AI has become a major FinOps subject,” according to Marc Trilling, CEO of Saegus. “Our most advanced clients are doubling or tripling an AI budget this year and the most conservative ones are increasing it by double digits. In a period of uncertainty, this is remarkable. If 2025 was characterised by an industrialisation of enthusiasm around AI, 2026 is the year of industrialising transformation and ROI.”
Massive deflation of a token price. The most visible component of AI is the cost of the token – as the visible cost of each prompt – which dropped down by 50% to 90% between 2024 and 2026 for major language models from OpenAI, Anthropic and Google. This trend is set to continue as these publishers launch new models.
At the end of Q1 2026, rates fluctuated between US$0.07 and US$0.3 per million tokens, compared to US$10 to US$30 two years earlier. “This upends the whole production economy: writing a product sheet or generating advertising banners now only costs a few cents,” according to Guilhem Bodin, Partner at Converteo. We heard the same observation at Tata Consultancy Services. “With US$1, companies go much further than 18 months ago,” shared Shekar Krishnan, VP & Head of Retail & CPG at TCS. “The gap between the cheapest and most powerful models is now exceeding x1,000.” For large models like GPT4o, Claude 3.5 Sonnet and Gemini 1.5 Pro, the price varied between US$10 and US$15 per million output tokens in early 2024. At the end of 2025, pressure from DeepSeek and xAI’s Grok (Elon Musk) caused mid-range prices to drop down by US$3 and US$12.
