DeepSeek V4.1 Flash vs Gemini 3.8 Flash

Both of these are cheap-tier models that outscore flagships from a year ago, and both come with a pricing asterisk you have to read. DeepSeek V4.1 Flash, released September 10, 2026, is a 552 billion parameter mixture-of-experts activating roughly 8 billion on input and 16 billion on output, with MIT weights on Hugging Face, a 1,048,576 token context, and native vision, and since September 14 it serves every DeepSeek API call, including requests still addressed to V4 Pro. It bills $0.30 per million input tokens and $1.20 output at peak, halving to $0.15 and $0.60 off-peak, where peak means Monday through Friday, 01:00 to 04:00 and 06:00 to 10:00 UTC. Gemini 3.8 Flash, released September 2, is $0.75 and $3.75 with the same 1,048,576 token window, but Google printed the expiry: that rate runs through December 31, 2026 and both sides double to $1.50 and $7.50 on January 1, 2027. So DeepSeek is cheaper today at every hour and roughly five times cheaper on output after the new year. On reported capability they are closer than the price gap suggests. DeepSeek puts GPQA Diamond at 90.9, Terminal-Bench 2.1 at 90.6, and DeepSWE v1.1 at 74.2, all self-reported; the agentic rows beat its own V4 Pro, though V4 Pro still edges it on GPQA Diamond at 92.4. Artificial Analysis independently measures Gemini 3.8 Flash at 95.3 on GPQA Diamond for the high reasoning tier, which is both a higher number and a better-sourced one. Google also takes text, image, audio, video, and PDF on one endpoint against DeepSeek's text and vision, and offers the operational maturity of Vertex and AI Studio. DeepSeek offers something Google structurally cannot: the weights, under MIT, with no rate card at all if you host them yourself.