Cheap open models win in practice
Many report DeepSeek Flash and similar models deliver far better price-performance for real coding work than Claude or OpenAI, especially with long caching and cheap subscriptions.
Methodology misses the real picture
Critics say general benchmarks and API list prices ignore task-specific coding results, subscription discounts, and local hardware costs, so the chart misleads.
Open weights matter beyond price
Some argue inspectability, customization, local running and freedom from usage restrictions make open models valuable regardless of where the frontier sits this month.