Story perspectives

DeepSeek's R1 Model: Costs Raise Training Debate

1/12/2026

Summary

Story summary

DeepSeek’s R1 model, launched in January 2025, sparked debate over its training costs.
Initial reports claimed a cost advantage over OpenAI's GPT-4, citing as little as $294,000 for post-training refinement.
However, total costs exceed $5.87 million for computing and $51 million in hardware, not including research and personnel expenses.
By contrast, OpenAI's GPT-4 training costs are estimated between $5 and $16 million, suggesting the gap may be smaller than claimed.

History

Historical Context

OpenAI launched GPT-3 in June 2020, a groundbreaking language model with 175 billion parameters, significantly enhancing AI capabilities. This innovation spurred competition among companies like DeepSeek to develop cost-effective AI solutions. The ongoing debate over training costs reflects the industry's drive to balance performance and affordability, shaping the future landscape of artificial intelligence.

1 of 2

Story summary

DeepSeek’s R1 model, launched in January 2025, sparked debate over its training costs.
Initial reports claimed a cost advantage over OpenAI's GPT-4, citing as little as $294,000 for post-training refinement.
However, total costs exceed $5.87 million for computing and $51 million in hardware, not including research and personnel expenses.
By contrast, OpenAI's GPT-4 training costs are estimated between $5 and $16 million, suggesting the gap may be smaller than claimed.

1 / 2