Overview
- DeepSeek opened a limited internal test of DeepSeek V4.1 Flash on Tuesday using the model id deepseek-v4.1-flash-expires-on-0910, which developers can call without changing their base_url and which is capped at 20 concurrent requests per account while billing matches the existing V4 Flash.
- The DeepSeek Open Platform announced a Flash-series price change on Wednesday that takes effect at 12:00 Beijing time on September 10, setting idle-period rates at input-hit ¥0.02 per million tokens, input-miss ¥1 per million, and output ¥4 per million with peak-period costs at twice those rates.
- DeepSeek describes V4.1 Flash as using a new model architecture with native multimodal support, faster inference and lower inference cost, but the company has not published technical details, parameter sizes or benchmark results for independent verification.
- DeepSeek asked testers whether the V4.1 Flash mid-version could fully replace the online V4 Pro, while keeping V4 Pro pricing unchanged, signaling a strategy to push Flash toward higher-tier performance at lower cost.
- Flash has been updated repeatedly this summer and the new test plus the price change could matter most to developers whose workloads have high cache hit rates because input-side costs fall more than output costs, but objective gains for Pro-style workloads remain to be proven by external testing.