Guides · Models
DeepSeek V4.1 Flash: price, context window and specs
| Developer | DeepSeek |
|---|---|
| Released | 10 September 2026 |
| Model ID | deepseek-flash (deepseek-v4-flash still accepted) |
| Parameters | 552B MoE; 8B active for input, 16B for output |
| Context window | 1M tokens (384K max output) |
| API price per 1M tokens (peak) | $0.30 input / $0.006 cache hit / $1.20 output |
| Off-peak price | $0.15 / $0.003 / $0.60 |
| Modes | Thinking (default) and non-thinking |
| Input / output | Text and image in; text out |
| Weights | Open, MIT license (Hugging Face) |
Tests by multiple parties put V4.1-Flash ahead of V4-Pro on performance, cost, speed & total runtime.
What DeepSeek says changed
DeepSeek's release note calls V4.1 Flash the smallest model in its new architecture family. It uses a causal encoder-decoder design with 8B active parameters for input and 16B for output, and DeepSeek says its KV cache needs a quarter of the HBM and an eighth of the SSD storage of the previous generation. The model card says it was trained from scratch on 45T multimodal tokens and processes images natively. DeepSeek says it beats its own flagship V4-Pro on benchmarks; that's the vendor's claim.
What happened to V4-Pro
DeepSeek says it's phasing out V4-Pro. From 04:00 UTC on 14 September 2026, all deepseek-v4-pro requests route to V4.1 Flash at V4.1 Flash rates, until V4.1-Pro launches. The older deepseek-v4-flash and deepseek-v4-flash-vision-exp names also route to V4.1 Flash. Note that DeepSeek's pricing page still lists a separate deepseek-v4-pro row (DeepSeek-V4-Pro-0813) with higher prices.
Pricing details
- Peak: $0.30 per million input tokens on a cache miss, $0.006 on a cache hit and $1.20 output.
- Off-peak rates are half the peak rates: $0.15 / $0.003 / $0.60.
- Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays. Every other hour is off-peak, including weekends.
- The new prices took effect at 04:00 UTC on 10 September 2026.
How to use it
- The API offers an OpenAI-format base URL (api.deepseek.com) and an Anthropic-format one (api.deepseek.com/anthropic), and supports the Responses API, JSON output and tool calls.
- Thinking mode is on by default; FIM completion works only in non-thinking mode.
- DeepSeek's docs include integration guides for Claude Code, Codex, OpenCode and other agents.
- The weights are on Hugging Face under the MIT license for self-hosting.
Independent benchmarks
- Artificial Analysis reported a score of 39 on its Intelligence Index (13 Oct 2026).
- Results from open datasets (Arena, Epoch AI) are on our benchmarks page.
Questions
- Is DeepSeek V4.1 Flash open source?
- The weights are published on Hugging Face under the MIT license. DeepSeek also serves the model through its own API as deepseek-flash.
- Does deepseek-v4-pro still work?
- The model name is still accepted, but DeepSeek says that since 14 September 2026 those requests are served by V4.1 Flash at V4.1 Flash prices until V4.1-Pro launches.
- When is the DeepSeek API cheapest?
- Off-peak, when prices are half the peak rate. Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, excluding Chinese public holidays.
Sources
- DeepSeek-V4.1-Flash release · DeepSeek API Docs ·
- Models and pricing · DeepSeek API Docs
- DeepSeek-V4.1-Flash model card · Hugging Face · deepseek-ai
- Artificial Analysis: DeepSeek V4.1 Flash model page · Artificial Analysis
Last checked . Facts change; the sources above are authoritative. See how we verify or report a correction.
