DeepSeek V4.1 Flash model cuts AI costs by 90% with high-performance benchmark

Precedent & Market Context

DeepSeek's latest model, V4.1-Flash, is the successor to DeepSeek-V4-Pro, which has largely been retired in favor of the new model. The company's use of the MoE architecture, which activates only a small slice of the network for each request, has been central to its strategy of offering a large model without paying the full compute cost every time.

Main Event

On October 17, 2026, DeepSeek pushed an update, labeled V4.1-Flash-1017, under the same API, which improved performance metrics and cut task runtimes. Independent tests show that V4.1-Flash still trails leading rivals on some of the most complex reasoning tasks, but excels in coding and agentic work.

Analysis & Market Impact

DeepSeek charges $0.15 per million input tokens and $0.60 per million output tokens during off-peak hours, with rates doubling during peak usage times. Cached inputs get even cheaper, with costs falling to between $0.003 and $0.006 per million tokens. This cache-hit pricing is approximately 30 times lower than certain frontier closed models.

Implications & Outlook

The availability of V4.1-Flash under the MIT license means that enterprises can self-host it, avoiding API dependency entirely. This flexibility matters for companies with data residency rules or a general reluctance to send sensitive information to an outside provider. However, the actual costs depend heavily on when and how a team uses the model, with peak-hour pricing doubling the headline rates.