Released on 24 April, DeepSeekV4 Flash sets a very low price threshold for advanced artificial intelligence. The model costs $0.14 per million input tokens and $0.28 per million output tokens. By comparison, GPT-5.6 Sol costs $5 and $30 respectively, whilst Claude Fable 5 costs as much as $10 and $50. Calculated per task completed, the difference can exceed a hundredfold, although this depends on the length of the response and the number of operations required to obtain the result.
However, the low price does not imply full equivalence. Artificial Analysis awards V4 Flash 40 points on its quality index, whilst OpenAI and Anthropic’s flagship models score around 59 points. DeepSeek, on the other hand, offers one million context tokens, fast response generation and open weights under the MIT licence. The model currently supports text but does not analyse images.
V4 Flash may be particularly attractive for high-volume, repetitive tasks such as customer service, document analysis, information retrieval or developer support. At a time when boards are assessing the return on investment in AI with increasing rigour, lower costs may facilitate the transition from pilot schemes to full-scale deployments.
The most significant consequence will be pressure on prices across the entire market. Providers of more expensive models will have to justify their pricing more strongly on the basis of quality, security and reliability. Companies are likely to start combining solutions: cheaper models will handle simple processes, whilst the most advanced systems will be reserved for tasks where errors are costly. Growing competition may accelerate the adoption of AI, but it will also heighten the importance of data governance, regulatory compliance and independent testing.
