DeepSeek V4.1 Flash is the most used model on OpenRouter right now: about 120 trillion tokens in the last 30 days. DeepSeek's own price is $0.30 per million input tokens and $1.20 per million output tokens, with a 1M-token context window. At least seven companies sell it at a standard price, and there are batch, off-peak and promotional rates on top. On 16 October the cheapest standard price was about 75% below DeepSeek's own.
Open the DeepSeek V4.1 Flash cost calculator or see every DeepSeek model and its prices
The short answer
- Lowest price: Cheaper Inference, a discount marketplace, at roughly a quarter of the list price. Nobody has published tests of whether what it serves matches DeepSeek's own API, so try it on your own prompts before moving real traffic.
- Lowest price from DeepSeek itself: its off-peak rate, half the list price outside peak hours. Peak is 01:00-04:00 and 06:00-10:00 UTC on weekdays; Chinese public holidays are off-peak all day. For work you can schedule, this is the cheapest option with no middleman.
- If answers can wait up to 24 hours: OpenRouter's batch price, 60-70% below list.
- Agents with long, repeated context: compare the cached-input price, not the headline price. It can matter more than everything else.
Every seller right now
| Seller | Input / 1M | Output / 1M | Cached / 1M |
|---|---|---|---|
| Cheaper InferenceDiscount marketplace | $0.074 | $0.30 | $0.0015 |
| RunInfraInference host | $0.14 | $0.58 | $0.030 |
| OpenRouterAggregator | $0.036 | $0.68 | $0.0054 |
| OrcaRouterAggregator · no cache discount | $0.15 | $0.60 | - |
| TokenRouterAggregator | $0.18 | $0.72 | $0.0036 |
| DeepSeekOfficial price | $0.30 | $1.20 | $0.0060 |
| Together AIInference host | $0.30 | $1.20 | $0.0060 |
| OpenRouter · BatchBatch: answers within 24 hours | $0.11 | $0.34 | $0.0034 |
| InferX · PromoPromo: 70% off its usual price, can end any time | $0.090 | $0.36 | $0.0018 |
| DeepSeek · Off-peakOff-peak: peak 01:00 - 04:00 and 06:00 - 10:00 UTC, Mon-Fri except Chinese public holidays; all other hours off-peak | $0.15 | $0.60 | $0.0030 |
USD per 1M tokens, last checked 16 Oct, 06:19 UTC. Batch, off-peak and promo prices are listed under their seller with what they mean.
What it costs per month
| Seller, per month | Chatbot | Coding agent | Document Q&A |
|---|---|---|---|
| Cheaper Inference | $59.26 | $191 | $94.94 |
| InferX · Promo | $71.79 | $231 | $115 |
| OpenRouter · Batch | $76.05 | $250 | $134 |
| OpenRouter | $94.15 | $299 | $98.82 |
| RunInfra | $118 | $436 | $205 |
| DeepSeek · Off-peak | $120 | $386 | $192 |
| OrcaRouter | $140 | $756 | $324 |
| TokenRouter | $144 | $463 | $230 |
| DeepSeek | $239 | $771 | $383 |
| Together AI | $239 | $771 | $383 |
10,000 requests a day, 30 days. Chatbot: 1,500 in / 400 out tokens, 30% cached · Coding agent: 12,000 in / 1,200 out tokens, 70% cached · Document Q&A: 6,000 in / 300 out tokens, 50% cached. Cheapest per column in bold.
On 16 October, a chatbot at 10,000 requests a day came to about $59 a month at the cheapest seller against $239 at DeepSeek's list price; a coding agent with 70% of its input cached came to $191 against $771. Use the calculator on the model page for your own token counts.
The catches
The cheapest seller is a discount marketplace
Cheaper Inference sells DeepSeek V4.1 Flash at around 75% below list. That is a lot, and the reasons a reseller can be that cheap range from bulk capacity to a smaller or compressed model behind the same name. Until there is public testing, treat it as something to verify: send it the same prompts as DeepSeek's API and compare the answers. If you have used it, a short review helps the next person.
OpenRouter's price is lopsided
On 16 October OpenRouter listed $0.036 per million input tokens but $0.684 per million output tokens: very cheap to read, close to list to write. OpenRouter routes each request to one of several hosts, so check which host and price applied in your activity log before planning around it. Its batch price is the better deal if you can wait.
Cache discounts differ more than prices do
OrcaRouter is cheaper than TokenRouter on headline price, but it publishes no cached-input price. For a coding agent that resends the same 12,000-token context with 70% of it cached, that difference made OrcaRouter about $756 a month against $463 at TokenRouter. If your workload repeats context, sort sellers by the cached price first.
Promos can end any day
InferX sells it at 70% off its usual rate as a promotion. That is real money while it lasts, but do not build a budget on it.
Which one should you pick?
- Starting out or low volume: DeepSeek's own API. At small volume the difference is a few dollars a month, and you get the reference version of the model.
- Scheduled or batch work: DeepSeek off-peak, or OpenRouter batch. Both roughly halve the bill.
- High volume and price-sensitive: test the cheapest sellers on a sample of your own prompts against DeepSeek's API first, then switch.
- Long-context agents: pick by cached-input price.
Prices on this page come from 13 public sources and refresh every 6 hours; the tables above update with them. How they are collected is on the method page.