LLM prices / DeepSeek / DeepSeek Flash
DeepSeek API · prices checked Oct 23, 2026
DeepSeek Flash API pricing
Short answerDeepSeek Flash costs $0.3 per 1M input tokens and $1.2 per 1M output tokens on the DeepSeek API (standard price, checked Oct 23, 2026). A typical chatbot reply (1,500 tokens in, 400 out) costs $0.0009, so 1,000 replies a day is about $27.90 a month and 1M replies is $930. That makes it #4 cheapest of the 29 current models for that job.
Best for
- High-volume simple jobs (support chat, tagging, extraction): #4 cheapest of 29 current models for a chatbot reply.
- Long repeated prompts (agents, RAG with a fixed prefix): cached input bills at 2% of the normal input price.
- Whole codebases or contracts in one prompt: 1M tokens of context.
- Reading more than text: takes image input.
Skip if
- You run large offline jobs: no batch price is listed.
Rules: Best for = cheapest quarter of current models for a chatbot reply, top 3 on one of our 8 jobs, cache reads at 15% or less of input, batch at 60% or less, 1M+ context, non-text inputs. Skip if = not current on the provider's page, priciest quarter, text only, no cache or batch price, a long-prompt surcharge, preview, under 200K context, no tool calling. Price and spec facts only; we don't score answer quality.
What your volume costs
Pick a job and a monthly volume. Same prices as everything on this page; caching and batch applied where they make it cheaper.
Cost per task, 8 jobs
Per 1,000 requests and per month at a typical volume, billed the cheapest way the price list allows. The rank is among current models that fit the job.
| Job (links to the cheapest model for it) | Per 1,000 | A month | Rank among fits |
|---|---|---|---|
| Chatbota 1,500-token prompt (system prompt plus the conversation so far) and a 400-token reply | $0.93standard price | $27.90at 30,000 a month | #4 of 29 |
| RAG (answer from your docs)6,000 tokens of retrieved passages plus the question, and a 500-token answer | $2.40standard price | $72.00at 30,000 a month | #5 of 29 |
| Coding agent40,000 tokens of repo context and history, 30,000 of them repeated from the last step and billed at the cache-read price, 2,000 tokens out, tool calling required | $5.58with the repeated 30,000 tokens billed at the cache-read price | $16.74at 3,000 a month | #4 of 29 |
| Tool-using agent12,000 tokens of instructions, tool definitions and history, 10,000 of them repeated from the last step and cached, 600 tokens out, tool calling required | $1.38with the repeated 10,000 tokens billed at the cache-read price | $13.80at 10,000 a month | #4 of 29 |
| Vision (read an image)one image counted as 1,000 input tokens plus a 300-token question, and a 300-token answer. Providers count image tokens differently, so treat vision costs as a rough guide | $0.75standard price | $7.50at 10,000 a month | #4 of 28 |
| Translationa 1,000-token passage in and about 1,100 tokens out (translations run a little longer than the source) | $1.62standard price | $16.20at 10,000 a month | #4 of 29 |
| Batch extraction2,000 tokens per record in and 300 tokens of JSON out, through the batch API where there is one (results within 24 hours) | $0.96standard price | $96.00at 100,000 a month | #7 of 29 |
| Summarize long documentsa 20,000-token document in and an 800-token summary out | $6.96standard price | $6.96at 1,000 a month | #5 of 29 |
Cheaper and pricier alternatives
The current models just below and just above DeepSeek Flash on the cost of 1,000 chatbot replies, with what differs. Price only, not quality.
Cheaper
- GPT-5.6 Luna $0.780.84x the cost of a chatbot reply, OpenAI, 1.05M context vs 1M, has a batch price.
- Claude Haiku 5.5 $0.350.38x the cost of a chatbot reply, Anthropic, has a batch price.
- GPT-6 Luna $0.350.38x the cost of a chatbot reply, OpenAI, 1.05M context vs 1M, has a batch price.
Pricier
- Gemini 3.1 Flash-Lite $0.971.05x the cost of a chatbot reply, Google, 1.05M context vs 1M, takes image, video, audio, pdf too, has a batch price.
- Gemini 3.5 Flash-Lite $1.451.56x the cost of a chatbot reply, Google, 1.05M context vs 1M, takes image, video, audio, pdf too, has a batch price.
- Gemini 3 Flash Preview $1.952.10x the cost of a chatbot reply, Google, 1.05M context vs 1M, takes image, video, audio, pdf too, has a batch price.
Every price tier
| Input | $0.3 per 1M tokens |
| Output | $1.2 per 1M tokens |
| Cache read (hit) | $0.006 (2% of the input price) |
| Offpeak tier | $0.15 input / $0.6 output |
| Also billed as | deepseek-v4-flash, deepseek-v4-flash-vision-exp |
| API model id | deepseek-flash |
| Status on the pricing page | listed |
Provider notes: input/output/cache_read are peak-hour prices; other_tiers.offpeak is the off-peak price (peak hours: 01:00 - 04:00 and 06:00 - 10:00 UTC, Monday through Friday, excluding Chinese public holidays); deepseek-v4-flash and deepseek-v4-flash-vision-exp are retired names now billed at the flash price; null means the page lists no price: batch_input, batch_output
Model facts from DeepSeek
| Context window | 1,000,000 tokens |
| Max output | 384,000 tokens |
| Input types | text, image |
| Tool calling | yes |
| Lifecycle | current |
| Provider note | Page states context length 1M and max output maximum 384K; served by DeepSeek-V4.1-Flash-1017; Vision supported. |
Source: https://api-docs.deepseek.com/quick_start/pricing/, checked Oct 24, 2026.
The 6 classic workloads
Per 1,000 requests at the standard price. "With caching" bills the repeated part of the prompt at the cache-read price (cache writes not counted); "via batch" uses the batch API price. Run your own token counts in the LLM API cost calculator.
| Workload | DeepSeek Flashper 1,000 requests |
|---|---|
| Chatbot replya 1,500-token prompt (1,000 of it a fixed system prompt) and a 400-token answer | $0.93$0.64 with caching |
| Answer from your docs (RAG)6,000 tokens of retrieved context plus the question, a 500-token answer | $2.40$2.11 with caching |
| Summarize a long documenta 20,000-token document in, an 800-token summary out | $6.96$6.67 with caching |
| Coding agent step40,000 tokens of repo context and history (30,000 of it repeated from the last step), 2,000 out | $14.40$5.58 with caching |
| Batch extraction2,000 tokens per record in, 300 tokens of JSON out, sent through the batch API where there is one | $0.96 |
| Long-document Q&Aa 150,000-token contract or codebase in one prompt, a 1,000-token answer | $46.20 |
What the numbers say
- Output costs 4x input on DeepSeek Flash, so for answer-heavy jobs the output price drives the bill; for document-heavy jobs (RAG, summaries, agents) input does.
- Caching matters: a coding-agent step that rereads 30,000 cached tokens costs 61% less than the same step uncached ($5.58 vs $14.40 per 1,000 steps, cache writes not counted). How to stop paying twice for repeated prompts: LLM semantic cache.
- Among 2 listed DeepSeek models it is the cheapest for a chat reply.
Every listed DeepSeek model, cheapest first
Same chatbot workload, 1,000 replies.
| Model | Input / 1M | Output / 1M | 1,000 chat replies |
|---|---|---|---|
| DeepSeek Flash | $0.3 | $1.2 | $0.93 |
| DeepSeek V4 Pro | $1.32 | $3.96 | $3.56 |
Closest price from other providers
The current model from each other provider whose chatbot-reply cost is nearest. Price only: we do not rank quality here.
| Model | Input / 1M | Output / 1M | 1,000 chat replies | vs DeepSeek Flash |
|---|---|---|---|---|
| Claude Haiku 5.5Anthropic | $0.1 | $0.5 | $0.35 | 0.38x |
| GPT-5.6 LunaOpenAI | $0.2 | $1.2 | $0.78 | 0.84x |
| Gemini 3.1 Flash-LiteGoogle | $0.25 | $1.5 | $0.97 | 1.05x |
Price history
- Oct 23, 2026: DeepSeek Flash added to the table: $0.3 input / $1.2 output per 1M tokens.
- Checked against the provider page on 1 date: Oct 23, 2026.
Every change we record is on the LLM price changelog.
Sources
- DeepSeek API pricing: https://api-docs.deepseek.com/quick_start/pricing, checked Oct 23, 2026.
- Context, inputs and lifecycle: https://api-docs.deepseek.com/quick_start/pricing/, checked Oct 24, 2026.
- Dataset: llm-prices (every row names its page and check date) and /llm/prices.csv.
Prices change often. If one is out of date, email matt@tensorust.com and we will fix it.