DeepSeek Introduces One of the World’s Cheapest Language Models

Price competition in the artificial intelligence market has entered a new phase. China’s DeepSeek has released its new V4-Flash language model, attracting the attention of specialists worldwide. According to ixbt.com, the development has become one of the world’s cheapest large language models in terms of test-task costs, enabling a sharp reduction in expenses. This was reported by Ixbt.com reports .

According to the research company Artificial Analysis, completing a single test task with DeepSeek V4-Flash costs an average of $0.03. This figure shows a record-low price compared with competing products. For example, the same task costs $3.15 with Anthropic’s Claude Fable 5 model, meaning the new Chinese model is more than 100 times cheaper in this regard.

Pricing Policy and Computing Costs

Experts point out that simply comparing token prices does not always produce an accurate result. To assess actual costs, the amount of computing required by the model to generate a response also plays an important role. The new model’s token pricing is also highly affordable: $0.14 per 1 million input tokens and $0.28 per 1 million output tokens.

For comparison, a single test costs $0.86 with Moonshot AI’s Kimi K3 model, while the figure reaches $1.86 with OpenAI GPT-5.6 Sol. This shows that DeepSeek’s new solution has become one of the market leaders in cost optimization.

Capabilities and Quality Metrics

Initially available in preview format, DeepSeek-V4-Flash and its Pro version retained their previous architecture and size after the official release. Notably, the developers only restarted the post-training process.

According to the Intelligence Index ranking, V4-Flash scored 50 out of 100 across nine tests covering programming, logical tasks, and office scenarios. This result is on par with Google Gemini 3.6 Flash. However, it trails Kimi K3, which scored 57 points, as well as Claude Opus 5, Claude Fable 5, and OpenAI GPT-5.6, which achieved even higher results.

Market Competition and Future Plans

As a reminder, DeepSeek became globally famous after releasing its R1 model in early 2025. The company then faced pressure from strong competitors such as Moonshot AI, MiniMax, Z.AI, ByteDance, and Alibaba. The company is currently preparing the more powerful V4-Pro model, but its release date remains undisclosed.

At the same time, Alibaba has further intensified competition in China’s artificial intelligence market by introducing its new flagship Qwen3.8-Max model. In these conditions, the affordable and efficient solutions presented by DeepSeek will help the company maintain and strengthen its position.