IBM CEO: Open-weight AI models now cost 90% less, driving enterprise shift
IBM CEO Arvind Krishna stated that open-weight AI models now achieve 90% of frontier model performance at 10-25% of the cost. This economic shift is driving significant enterprise demand for on-premises AI infrastructure. Krishna noted that while closed models led in 2024, the performance gap has narrowed substantially, making cheaper open alternatives viable for most business tasks like email and contract analysis.
Key points
- Arvind Krishna, CEO of IBM, reported that open-weight AI models now deliver 90% of the effectiveness of closed frontier models.
- Krishna stated that open models cost only 10% to 25% of the token price of frontier models, creating a strong economic incentive for enterprises.
- The performance gap between top closed models and open alternatives has narrowed from a 70-to-30 score differential in 2024 to 70-to-50 in late 2025.
- IBM is experiencing increased interest in on-premises AI infrastructure, with Krishna noting that most businesses can deploy useful systems using just two or three racks.
- Krishna argued that most routine enterprise tasks, such as summarizing emails or answering customer calls, do not require the most expensive frontier models.
- Despite the cost advantage, Krishna acknowledged that some specialized tasks still require higher-cost frontier models, though the majority of use cases favor open alternatives.
IBM CEO Arvind Krishna has declared that open-weight artificial intelligence models have reached a critical economic threshold, making them highly attractive for enterprise deployment. Speaking during the company's Q3 earnings call, Krishna highlighted that these open models now achieve approximately 90% of the performance of top-tier closed frontier models while costing only 10% to 25% as much in token expenses. This dramatic cost-performance ratio is fundamentally altering how businesses approach AI infrastructure.
Krishna noted that in 2024, closed models held a distinct advantage, with benchmarks showing top models scoring around 70 compared to 20 or 30 for open alternatives. By late 2025, that gap has narrowed significantly, with open models now scoring around 50. For most routine business operations—including email summarization, contract examination, and customer service—this level of performance is sufficient, leading Krishna to predict that enterprises will increasingly opt for the cheaper open-weight options.
This shift is driving a surge in demand for on-premises AI hardware. Krishna indicated that while many companies can build functional AI infrastructure with just two or three server racks, the complexity of deployment remains a challenge. As the cost barrier drops, the competitive landscape for AI providers is evolving, with open-source ecosystems gaining substantial traction in the corporate sector.
Sources
The WireByte editorial team synthesises technology news from multiple primary sources, verifies the facts, and links every source. Articles are produced with AI assistance and reviewed under our editorial policy.
