Guides · dev · openrouter deepseek v4 flash

DeepSeek V4.1 Flash on OpenRouter: the cheap default, with the direct price beside it

Published 16 Oct 2026. Updated 16 Oct 2026. Researched 16 Oct 2026.

If you want one Chinese id as the default cheap workhorse, V4.1 Flash is the one we can tie to a vendor price page. DeepSeek’s docs say the API name deepseek-flash serves DeepSeek-V4.1-Flash. Older names still resolve and are billed at the flash price. That is the direct API. OpenRouter’s id is deepseek/deepseek-v4.1-flash.

The two prices

Peak, direct, per million tokens: $0.30 input on a cache miss, $1.20 output, $0.006 on a cache hit. Off-peak is half. OpenRouter’s list price on this snapshot matches the peak miss and the peak output, not the off-peak pair.

There is also deepseek/deepseek-v4-flash on OpenRouter, which is cheaper still. DeepSeek’s pricing page says the legacy flash names are served by V4.1 Flash and billed at the V4.1 rate. So a cheaper OpenRouter card with the old name may be a different host serving a different checkpoint. Do not assume the old id is “the same model, half off”.

Three DeepSeek cards, not three copies of one price. OpenRouter list price, blended 75% input / 25% output. Snapshot 2026-10-16 06:10 UTC. Comparators resolve to Anthropic: Claude Opus 5.5 and OpenAI: GPT-6.1 Sol.
Model	Blended	vs Opus	vs Sol
deepseek-v4.1-flash	$0.198	98% cheaper	95% cheaper
deepseek-v4-flash	$0.114	99% cheaper	97% cheaper
deepseek-v4-pro	$1.19	85% cheaper	70% cheaper

Calling it

On OpenRouter you use your OpenRouter key and the id above. On DeepSeek you use https://api.deepseek.com and the model name deepseek-flash. The Anthropic-compatible base is https://api.deepseek.com/anthropic. We did not run a secret-bearing prompt through either.

Coding-agent session: 3 million input tokens, 90% served from cache, 60 thousand output tokens. Cache-write fees are not included. Chinese rows use the lab published rate (DeepSeek uses the peak rate). Comparators are OpenRouter list prices for Anthropic: Claude Opus 5.5 ($8.00/M blended) and OpenAI: GPT-6.1 Sol. Snapshot 2026-10-16. Cache-write fees are excluded. A missing cache price is billed as a full input token.
Model	Session	vs Opus	vs Sol	Cache price
DeepSeek-V4.1-Flash	$0.178	94% cheaper	88% cheaper	published
Anthropic: Claude Opus 5.5	$2.94	—	—	published
OpenAI: GPT-6.1 Sol	$1.47	—	—	published

The percentages in this table compare session cost. The price table's percentages compare blended token price. They answer different questions.

That is the peak direct rate on our illustrative session. Off-peak would be about half, because DeepSeek’s rule is a straight half, not a mystery discount. Cache-write fees are still excluded. If your agent does not actually hit cache, delete the 90 percent assumption before you believe the number. The calculator lets you switch to a short chat with no cache.

When not to default to it

Long private-repo agents, anything a government policy forbids, and any job where the missing V4 row on SWE-bench Verified makes you nervous. The comparison says that more carefully. For bulk classification and drafts, this is the id we would try first.

Questions

Which OpenRouter id is the current flash model?

deepseek/deepseek-v4.1-flash. The older deepseek-v4-flash id is still listed and is a different, cheaper card. DeepSeek's own docs say the API name deepseek-flash now serves V4.1 Flash.

Is there a free DeepSeek route on OpenRouter?

Not in the priced rows we keep. Zero-price variants are dropped by the build. A :free suffix would be omitted on purpose.

Sources

DeepSeek API pricing, accessed 2026-10-16
OpenRouter models API, accessed 2026-10-16
DeepSeek V4.1 Flash note, accessed 2026-10-16