Artificial intelligence (AI) models could soon become more expensive, forcing businesses to reassess their AI investment strategies.
Anthropic’s latest AI model costs four times more than before but outperforms its previous top model. The new Claude 3.5 Haiku charges $1 to process a million tokens of input and $5 to generate a million tokens of output. Experts say the escalating costs of high-performance AI models can pose significant challenges for small- to medium-sized businesses (SMBs).
“Developing custom AI solutions can be expensive, with costs ranging from $20,000 to over $500,000, depending on complexity,” Niv Hertz, director of AI at Aporia, told PYMNTS. “This financial barrier may limit SMBs’ ability to adopt advanced AI technologies, potentially widening the competitive gap between them and larger enterprises.”
The new Claude 3.5 Haiku improves upon its cheaper predecessor and even outperforms Anthropic’s flagship Claude 3 Opus model. Although it costs more than the original Haiku, it’s still more affordable than Claude’s Opus, which charges $15 input and $75 output per million tokens.
For enterprise users, OpenAI’s GPT-4 costs around $60 per million tokens for the 8k model and about $120 per million tokens for the more advanced 32k context. Microsoft’s Azure OpenAI Service offers similar pricing, with GPT-4 8k model access at roughly $60 per million tokens.
A token in AI language models represents a single unit of text, such as a word or part of a word, which the model uses to process and generate responses.
With increasing costs for advanced AI models, SMBs will likely turn to small language models (SLMs), which tend to be more finely tuned, Raj Koneru, CEO and founder of Kore.ai, told PYMNTS.
“These models, with fewer parameters, offer a cost-effective solution tailored for specific industry applications like banking, healthcare and retail, enabling faster, more accurate responses,” he added. “SLMs provide an accessible AI alternative for businesses needing conversational capabilities without the cost of large-scale models.”
Koneru said companies have options tailored to specific industries like healthcare and HR as AI becomes more specialized. This targeted approach is reshaping how AI services are priced and delivered.
“As a result, businesses can select AI solutions that fit their industry’s unique requirements, setting a new standard for accessible, customizable and cost-effective AI functionalities across sectors,” he said.
AI costs are becoming more affordable, not more expensive, NetMind.AI Chief Commercial Officer Seena Rejal told PYMNTS. While top-tier models may command premium prices, comparable performance to last year’s costly models is now available at lower costs. Companies like OpenAI and Anthropic have consistently reduced prices for their AI capabilities, making powerful AI more accessible to businesses of all sizes.
Rejal said that as AI models become more specialized, we will likely see the emergence of new pricing tiers that reshape competition and influence consumer expectations, particularly in sectors like eCommerce and media. This shift is driven by the need for tailored solutions that address specific industry challenges and customer needs.
“The rise of specialized AI offerings is leading to the adoption of tiered pricing models,” he said. “Businesses increasingly recognize the value of AI models tailored to niche applications, enabling value-based pricing strategies. Solutions that address specific challenges with measurable benefits can command premium prices, as companies are willing to pay more for tools that directly solve their unique needs.”
According to K2view’s Chief Technology Officer Yuval Perlov, companies want better value from their AI — specifically, higher accuracy without high costs. He told PYMNTS that many firms are turning to Retrieval Augmented Generation (RAG) frameworks, which combine two fundamental processes: retrieving specific information from company databases and augmenting AI responses. This approach boosts both accuracy and personalization without having to conduct expensive retraining for AI models.
“The AI landscape is rapidly changing,” Fang Chang, Chief Product Officer at Coupa Software, told PYMNTS. “Competition among industry leaders increases the development of AI across various sectors. Rather than impairing small businesses, these constant innovations help drive the cost of AI down, ultimately making the technology more accessible to companies with tighter budgets.”