A new price floor: $0.15/$0.60 per 1M off-peak, doubling at peak hours, with native multimodal input and a 1M window.
DeepSeek released V4.1-Flash on September 10, 2026, retiring both V4-Flash and the experimental V4-Flash-Vision-Exp variant — the deepseek-v4-flash and deepseek-v4-flash-vision-exp API names now route to it for compatibility. It adds native multimodal (text + image) support, keeps the full 1M context window (384K max output), and cuts the off-peak rate to $0.15 input / $0.60 output per 1M tokens, down from V4-Flash's $0.22/$0.66. Peak hours (01:00-04:00 & 06:00-10:00 UTC weekdays) still double the rate, to $0.30/$1.20.
DeepSeek had planned to also route the deepseek-v4-pro alias to this same model from September 14, 2026, pending a future V4.1-Pro release — but reversed that plan before it took effect, in response to user demand. V4-Pro keeps its own, higher price ($0.66/$1.98 off-peak) as a separate, heavier reasoning-tier model.
The considerations are unchanged and non-technical: the hosted API routes through Chinese infrastructure, a hard blocker for some compliance regimes, and open-weights self-hosting is part of DeepSeek's pitch for teams that need to avoid it. If your workload is flexible on timing, scheduling batch jobs outside the peak window keeps costs at the lower rate.
For the price class, capability is still absurd: usable reasoning on math, code, and logic with selectable thinking-effort levels (low/high/max), a 1M window, native image input, and throughput suited to volume pipelines. Most tasks that teams route to mini-tier US models run fine here at a fraction of the cost, even at the new peak rate.
The honest weakness: polish and ecosystem. Tooling, SDK maturity, rate-limit headroom, and English prose quality all trail the US providers. Billing now also requires tracking time-of-day, adding a small layer of planning that flat-rate rivals don't need.
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| DeepSeek V4.1-Flash (off-peak) | $0.15 | $0.60 | 1M |
| DeepSeek V4.1-Flash (peak) | $0.30 | $1.20 | 1M |
| DeepSeek V4-Pro (off-peak) | $0.66 | $1.98 | 1M |
| GPT-5.4 nano | $0.20 | $1.25 | 400K |
| Mistral Small 4 | $0.15 | $0.60 | 256K |
Off-peak, V4.1-Flash now matches Mistral Small 4's per-token price exactly but with a 1M window versus Mistral's 256K, and it undercuts GPT-5.4 nano's input price while its 400K window is under half of V4.1-Flash's. If Chinese hosting is acceptable (or you self-host), V4.1-Flash off-peak is the rational default for cheap volume; if not, GPT-5.4 nano and Mistral Small are the compliant runners-up.