DeepSeek V4.1 Flash API
| Provider | Input / 1M | Cached / 1M | Output / 1M |
|---|---|---|---|
| Morph | $0.12 | $0.008 | $0.80 |
| DeepInfra | $0.14 | $0.0042 | $0.42 |
| Nebius | $0.30 | $0.01 | $1.20 |
| Novita AI | $0.30 | $0.006 | $1.20 |
| Inference.net | $0.30 | $0.007 | $1.00 |
| RunInfra | $0.14 | $0.03 | $0.58 |
| Sail Research | $0.09 | $0.006 | $0.30 |
USD, as of October 27, 2026. Live prices
| Workload | Morph | Sail Research |
|---|---|---|
| 1,000 chat replies2K in, 500 out each | $0.64 | $0.33 |
| 1,000 agent tasks50K in, 5K out each | $10.00 | $6.00 |
| Monthly batch job500M in, 50M out | $100.00 | $60.00 |
curl https://api.routerplus.com/v1/chat/completions \
-H "Authorization: Bearer $ROUTERPLUS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek/deepseek-v4.1-flash",
"messages": [{ "role": "user", "content": "Hello" }]
}'On RouterPlus, routing for DeepSeek V4.1 Flash starts at Morph: $0.12 per million input tokens and $0.80 per million output tokens. The lowest listed prices are at Sail Research: $0.09 input and $0.30 output. Name a provider with provider.only and you pay its price. Prices as of October 27, 2026.
7 providers serve DeepSeek V4.1 Flash on RouterPlus: Morph, DeepInfra, Nebius, Novita AI, Inference.net, RunInfra and Sail Research. By default the router tries them in routing order and fails over before the first byte, never mid-answer.
Use "deepseek/deepseek-v4.1-flash" as the model with the OpenAI-compatible endpoint at https://api.routerplus.com/v1. The same key also accepts the Anthropic API format.
1.05M tokens of context and up to 524K output tokens, per the RouterPlus catalog. A provider can list a lower limit; the console shows each one.
Yes. DeepSeek publishes DeepSeek V4.1 Flash as an open-weight model, which is why several independent providers can host it.
6 of 7 providers list no prompt retention (DeepInfra, Nebius, Novita AI, Inference.net, RunInfra and Sail Research). Morph may retain prompts. Pin the providers you accept with a routing policy.
Yes. Tell us the model, the throughput you need and the region with "I need inference", and RouterPlus lines up reserved capacity for it.
More DeepSeek