IBM CEO Krishna Says Open-Weight Models Perform Well and Cost Little
IBM CEO Arvind Krishna said the performance gap between open-weight models and frontier models has narrowed from 70 points versus 20 or 30 points in 2024 to 70 points versus 50 points by late 2025. He noted that the token cost of current open-weight models is only 10% to 25% of that of frontier models, while their effectiveness reaches 90% of frontier models. He believes most enterprises will therefore choose open-weight models for the vast majority of routine tasks and use frontier models only for a small number of tasks.
Why read this
AI score 56
IBM CEO compares benchmark scores and token costs to show that the cost-effectiveness of open-weight models for routine enterprise tasks is approaching that of frontier models.
Topics
Body · AI translation
IBM CEO Arvind Krishna said that deploying open-weight models within enterprises is extremely valuable. He noted that in 2024, closed-source frontier models significantly outperformed open-source options in performance. A frontier model might score 70 on benchmarks, while an open-weight model might score only 20 or 30. But by late 2025 the gap had narrowed—no longer 70 versus 20 or 30, but 70 versus 50. Today's open-weight models cost only 10% to 25% as much per token as frontier models, while achieving 90% of the effectiveness of frontier models. "For any enterprise, this is a simple question: if there is an option that costs only 10% of the top option and achieves 90% of its effectiveness, would you choose it, or would you be willing to pay 10 times the cost?" He believes most enterprises will choose to avoid bearing the extra cost. The vast majority of routine tasks can use open-weight models, while a small number of tasks can use frontier models.