DeepSeek Pricing 2026: Free Chat, App & API Costs
Independent guide — not an official DeepSeek pricing or sales page: Chat-Deep.ai is independent and is not affiliated with, endorsed by, or operated by DeepSeek. We do not set DeepSeek prices or sell official DeepSeek subscriptions, API keys, credits, or billing plans.
Quick answer: As of September 3, 2026, DeepSeek’s official website presents its web chat as free access, and its official Android listing presents the assistant as free. The official DeepSeek API is a separate, usage-based product billed by tokens. “Free” does not mean unlimited access, permanent terms, guaranteed availability, or a service-level agreement; limits and plans can change, so verify the current official pages before relying on them.
Current API price snapshot: The official matrix lists deepseek-flash (DeepSeek-V4.1-Flash, released September 10, 2026) and deepseek-v4-pro (DeepSeek-V4-Pro-0813). V4.1-Flash is listed at $0.003 / $0.15 / $0.6 off-peak and $0.006 / $0.3 / $1.2 at peak per 1M cache-hit input / cache-miss input / output tokens. Pro is $0.022 / $0.66 / $1.98 off-peak and $0.044 / $1.32 / $3.96 at peak. Images sent to V4.1-Flash are converted into billed input tokens, up to 1,024 per image. See the DeepSeek V4.1-Flash guide for specifications and migration.
Last verified by Chat-Deep.ai: September 18, 2026 (API models, prices, Vision guide and change log); September 3, 2026 (free web and app access).
Is DeepSeek Free?
It depends on the product. As of September 3, 2026, the official DeepSeek web chat and official app are presented as free to access. The official API is billed by tokens, while local use of open-weight releases can avoid a hosted API bill but still creates infrastructure and operating costs.
Free chat access is not a promise of unlimited messages, permanent pricing, uninterrupted availability, data residency, enterprise support, or an SLA. Check the official service and current terms for limits that apply to your account and region.
DeepSeek Free Chat vs DeepSeek API Pricing
Use this table to identify who operates each access path, who pays, and which limits matter.
| Access path | Operated by | Cost model | What to verify |
|---|---|---|---|
| Official DeepSeek web/app chat | DeepSeek | Presented as free access in the official sources checked on September 3, 2026 | Account, region, usage, availability, and current terms; free does not promise unlimited use or an SLA. |
| Official DeepSeek API | DeepSeek Platform | Usage-based token billing | Your organization pays for cache-hit input, cache-miss input, image input where applicable, and output tokens. |
| Open weights / local deployment | You or your infrastructure provider | A specific release may have no download or licence fee, but deployment is not cost-free | Exact model licence plus GPU or cloud compute, storage, electricity, security, monitoring, and maintenance. |
| Third-party apps and tools | The named third party | Provider-specific free tier, subscription, markup, or limits | Confirm the operator, model, privacy terms, quotas, and who handles billing. |
Does DeepSeek have a free API tier?
The public official pricing page checked for this guide does not document a permanent, universal free API tier. If the official platform shows a granted balance or promotional credit in your account, treat it as account-specific: check the amount, eligibility, expiry date, and applicable models. Do not build a production budget on an assumed free allowance.
Are DeepSeek open weights free to run locally?
Access to a particular model release may carry no download or licence fee, subject to that release’s exact licence. Running it still has a total cost of ownership: GPU or cloud time, storage, electricity, networking, security, monitoring, updates, and staff time. Review the DeepSeek system requirements and local deployment guide before comparing local cost with the hosted API.
DeepSeek API Pricing by Model
The official DeepSeek API pricing table is listed per 1 million tokens. The current official models are deepseek-flash (DeepSeek-V4.1-Flash) and deepseek-v4-pro (DeepSeek-V4-Pro-0813). V4.1-Flash has the lower rate on every line and accepts images natively; image tokens are billed as input tokens. The legacy name deepseek-v4-flash is still accepted and billed at Flash prices. DeepSeek first announced that V4-Pro requests would route to V4.1-Flash from September 14, 2026, then kept the V4-Pro API service running after that date with billing unchanged. The table lists the current off-peak and peak rates for each model.
| Rate period | Model | Cache-hit input / 1M | Cache-miss input / 1M | Output / 1M |
|---|---|---|---|---|
| Current off-peak | deepseek-flash | $0.003 | $0.15 | $0.6 |
| Current peak | deepseek-flash | $0.006 | $0.3 | $1.2 |
| Current off-peak | deepseek-v4-pro | $0.022 | $0.66 | $1.98 |
| Current peak | deepseek-v4-pro | $0.044 | $1.32 | $3.96 |
Pricing was checked against the official DeepSeek Models & Pricing page on . Peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday; every other hour, including all weekend hours, uses the off-peak rates. Peak rates are twice the current off-peak rates. Confirm the official page before production use.
DeepSeek converts each image sent to deepseek-flash into input tokens based on its dimensions and bills them at the normal input rate. The official Vision guide documents automatic resizing (small images are scaled up to about 544×544 pixels, large ones down to about 1300×1300) and a maximum charge of 1,024 input tokens per image; multiple images are counted independently. That image-token ceiling is not the total request cost because text input, conversation context, output, and repeated calls are billed separately. See the DeepSeek V4.1-Flash vision rules for details.
Check the official DeepSeek Models & Pricing page for the latest public API rates.
DeepSeek Price per 1K Tokens
DeepSeek publishes API prices per 1M tokens. Divide the applicable per-1M rate by 1,000 to estimate the price per 1K tokens. These are the current rates verified on September 18, 2026; apply the row for the model and UTC billing window used by the request.
| Model | Rate window | Cache-hit input / 1K | Cache-miss input / 1K | Output / 1K |
|---|---|---|---|---|
deepseek-flash | Off-peak | $0.000003 | $0.00015 | $0.0006 |
deepseek-flash | Peak | $0.000006 | $0.0003 | $0.0012 |
deepseek-v4-pro | Off-peak | $0.000022 | $0.00066 | $0.00198 |
deepseek-v4-pro | Peak | $0.000044 | $0.00132 | $0.00396 |
Example formula: (cache-hit input × hit rate) + (cache-miss input × miss rate) + (output × output rate). Images sent to V4.1-Flash add billed input tokens (up to 1,024 per image), so include image-derived tokens in the input total.
How DeepSeek API Billing Works
DeepSeek API billing is token-based, not a fixed price per request. A short request with a short answer costs less than a long request with a long answer. A request with a high cache-hit share can also cost less than one where every input token is processed as a cache miss.
| Billing line | What it means | Why it matters |
|---|---|---|
| Input tokens, cache hit | Input tokens served from DeepSeek context caching at the lower cache-hit price. | Repeated or reusable prefixes can reduce API costs when they receive cache-hit pricing. |
| Input tokens, cache miss | Input tokens that require fresh processing and are billed at the higher cache-miss input price. | New prompts, changed prefixes, and uncached context usually increase input cost. |
| Output tokens | Tokens generated by the model in the response and billed at the selected model’s output rate. | Longer answers cost more, so output limits are important for budget control. |
DeepSeek states that API expenses are deducted from topped-up balance or granted balance, with granted balance used first when both are available. For production planning, review the official rates, then use the DeepSeek API guide for integration details and model selection.
Cache Hit vs Cache Miss Pricing
DeepSeek context caching is enabled by default. A cache hit happens when a later request can reuse a persisted matching prefix from earlier requests. A cache miss happens when the input cannot be served from an already persisted matching prefix and requires fresh processing.
Do not assume every repeated prompt receives cache-hit pricing. DeepSeek describes context caching as best-effort, and cache hits depend on matching persisted prefixes. For accurate billing analysis, track prompt_cache_hit_tokens and prompt_cache_miss_tokens in the API response.
Listed rates show the formula; our measured cost of the DeepSeek 1M context benchmark shows dated spend derived from provider-reported hit, miss, and completion tokens across the complete controlled run.
| Usage field | Meaning | Cost impact |
|---|---|---|
prompt_cache_hit_tokens | The number of input tokens that received cache-hit treatment. | These tokens are billed at the lower cache-hit input rate. |
prompt_cache_miss_tokens | The number of input tokens that did not hit the cache. | These tokens are billed at the higher cache-miss input rate. |
completion_tokens | The number of generated output tokens. | These tokens are billed at the model’s output-token rate. |
When cache-hit pricing can help
- Multi-round conversations with a stable history prefix.
- Repeated system prompts that stay at the beginning of the request.
- Long static documents reused across several questions.
- RAG or chatbot workflows where the same context prefix is reused.
- Batch analysis where multiple requests share a stable instruction and document prefix.
When cache-hit pricing may not help
- Every request uses a completely new prompt or document.
- The reusable content changes position or no longer fully matches a persisted prefix.
- The workload has low repetition and few overlapping prefixes.
- You rely on cache-hit assumptions instead of measuring actual usage fields.
DeepSeek V4.1-Flash vs V4 Pro: Which Should You Choose?
Choose deepseek-flash (DeepSeek-V4.1-Flash) for most work: routine chat, summarization, extraction, classification, coding help, high-volume workloads, and requests that must analyze JPEG, PNG, GIF, or WebP images. It has the lowest listed token rate on every line, and each image adds billed input tokens (up to 1,024 per image). Disable thinking where a task does not need reasoning.
Evaluate deepseek-v4-pro on representative text tasks rather than assuming it is automatically better; DeepSeek’s own announcement reports V4.1-Flash ahead of V4-Pro. Pro costs more on every rate line and does not accept images, so choose it only when measured gains in accuracy, coding success, reliability, or task completion justify the higher rate for your workload.
| Use case | Recommended starting model | Reason |
|---|---|---|
| High-volume text chatbot | deepseek-flash | The lowest listed token rates; disable thinking when the task does not need reasoning. |
| Screenshot, chart, or image understanding | deepseek-flash | Native image input on the same model; budget up to 1,024 input tokens per image. |
| Summarization or extraction from text | deepseek-flash | Cost-sensitive text work at the lowest rate. |
| Complex coding task | deepseek-flash at high or max effort, then compare deepseek-v4-pro | Raise the reasoning effort before changing model; choose Pro only if the measured benefit justifies the total cost. |
| Agentic workflow | deepseek-flash, with deepseek-v4-pro as a candidate to test | Compare task completion, errors, latency, and total cost across candidate models. Pro is not an automatic quality upgrade. |
| Budget testing | deepseek-flash | One model for text and images at the lowest listed rates. |
Current DeepSeek API Model Details
The official DeepSeek Models & Pricing page currently lists deepseek-flash, serving DeepSeek-V4.1-Flash, and deepseek-v4-pro, serving DeepSeek-V4-Pro-0813. Both list OpenAI-compatible and Anthropic-compatible formats, thinking and non-thinking modes, JSON Output, Tool Calls, native Responses API support, a 1M-token context length, and a 384K maximum output. Only V4.1-Flash accepts image input. The retired V4-Flash and V4-Flash Vision Exp models are no longer listed; the legacy name deepseek-v4-flash is still accepted and served by V4.1-Flash.
| Feature | deepseek-flash | deepseek-v4-pro |
|---|---|---|
| Model version | DeepSeek-V4.1-Flash | DeepSeek-V4-Pro-0813 |
| Release status | Current; released September 10, 2026 | GA; API service continued after September 14, 2026 with billing unchanged |
| OpenAI-compatible base URL | https://api.deepseek.com | https://api.deepseek.com |
| Anthropic-compatible base URL | https://api.deepseek.com/anthropic | https://api.deepseek.com/anthropic |
| Image input | Supported: JPEG, PNG, GIF, WebP | Not supported |
| Maximum image-token charge | 1,024 input tokens per image | Not applicable |
| Context length | 1M tokens | 1M tokens |
| Maximum output | 384K tokens | 384K tokens |
| Thinking and non-thinking modes | Supported | Supported |
| JSON Output | Supported | Supported |
| Tool Calls | Supported | Supported |
| Responses API | Supported | Supported |
| Chat Prefix Completion | Supported, beta | Supported, beta |
| FIM Completion | Supported in non-thinking mode, beta | Supported in non-thinking mode, beta |
| Concurrency limit | 2500 | 500 |
FIM documentation note (matrix rechecked September 18, 2026; API reference last checked September 3, 2026): DeepSeek’s Models & Pricing matrix lists FIM Completion for deepseek-flash and deepseek-v4-pro in non-thinking mode only. The FIM guide uses Pro in its sample; a sample model does not by itself narrow the support matrix. However, the current /beta/completions API reference still enumerates only deepseek-v4-pro. Because those official pages are not fully aligned, use https://api.deepseek.com/beta, keep FIM output at or below 4K tokens, and validate V4.1-Flash against your account before production.
Legacy Compatibility Aliases (Historical)
The older model names deepseek-chat and deepseek-reasoner were legacy compatibility aliases. Before the announced retirement cutoff, DeepSeek mapped deepseek-chat to the non-thinking mode of deepseek-v4-flash and deepseek-reasoner to its thinking mode. Those mappings are historical migration context, not current documented model IDs.
Official alias status rechecked September 18, 2026; latest live alias test July 28: DeepSeek’s current official pricing matrix lists two explicit IDs—deepseek-flash and deepseek-v4-pro—notes that the legacy name deepseek-v4-flash is still accepted and served by DeepSeek-V4.1-Flash, and does not list deepseek-chat or deepseek-reasoner. Chat-Deep.ai observed the old aliases return HTTP 400 on July 25, then return HTTP 200 and route to V4 Flash in a bounded July 28 retest. That changing compatibility behavior does not create a separate model or pricing tier and is not a production guarantee. Use deepseek-flash for the behavior-preserving migration (text and images) and Pro as a separately evaluated higher-cost choice.
| Legacy name | Historical pre-cutoff mapping | Behavior-preserving replacement |
|---|---|---|
deepseek-chat | deepseek-v4-flash non-thinking mode. | deepseek-flash with thinking disabled. |
deepseek-reasoner | deepseek-v4-flash thinking mode. | deepseek-flash with thinking enabled. |
DeepSeek API Cost Examples
The examples below use the official rates verified on September 18, 2026. They exclude taxes, retries, third-party markups, and infrastructure costs. Actual billing depends on model, cache ratio, token counts, image input, and whether the request falls in the peak or off-peak UTC window.
| Scenario | Model | Token usage | Cache assumption | Off-peak | Peak |
|---|---|---|---|---|---|
| Simple chatbot reply | deepseek-flash | 1,000 input + 500 output | 0% cache hit | ~$0.00045 | ~$0.00090 |
| Pro comparison request | deepseek-v4-pro | 1,000 input + 500 output | 0% cache hit | ~$0.00165 | ~$0.00330 |
| Long document summary | deepseek-flash | 50,000 input + 5,000 output | 0% cache hit | ~$0.0105 | ~$0.0210 |
| Complex coding evaluation | deepseek-v4-pro | 20,000 input + 5,000 output | 0% cache hit | ~$0.0231 | ~$0.0462 |
| Cached repeated prompt | deepseek-flash | 100,000 cache-hit input + 5,000 output | 100% cache hit input | ~$0.0033 | ~$0.0066 |
| Cached Pro evaluation | deepseek-v4-pro | 100,000 cache-hit input + 5,000 output | 100% cache hit input | ~$0.0121 | ~$0.0242 |
| High-volume support bot | deepseek-flash | 1,000 input + 500 output × 30,000 requests/month | 0% cache hit | ~$13.50/month | ~$27.00/month |
Use the token usage fields returned by the API for billing analysis. For quick budgeting, use the DeepSeek API cost calculator, then verify final assumptions against the official pricing page.
Pricing Guidance and Official DeepSeek Services
This page explains free official DeepSeek chat access and official DeepSeek API billing. Chat-Deep.ai is a separate, independent website; it is not the official DeepSeek chat and does not sell official DeepSeek API credits, keys, subscriptions, or billing plans.
Chat-Deep.ai publishes independent guides, examples, tests, and pricing explainers. It does not provide a chat service. Developers use the official DeepSeek Platform for official API keys, balance, usage, and production billing.
| Topic | Read our independent guidance | Use official DeepSeek services |
|---|---|---|
| Understand browser chat options | Read our DeepSeek introduction to compare the official chat, API, and model releases. | Use the official DeepSeek website or app for the official consumer experience. |
| API keys and billing | Chat-Deep.ai does not sell official DeepSeek API keys, credits, or plans. | Use the official DeepSeek Platform for API keys, balance, usage, and billing. |
| Production applications | Use our API guide to understand workflows, terminology, and model choices. | Use official DeepSeek API docs and official pricing for production decisions. |
How We Verify DeepSeek Pricing
We verify DeepSeek API rates against the official DeepSeek Models & Pricing page, then convert the official per-1M-token rates into per-1K-token and example request estimates. We separate cache-hit input, cache-miss input, and output tokens because DeepSeek bills them at different rates.
We also check DeepSeek’s official token usage, context caching, rate limit, and FAQ documentation when explaining billing behavior. Because DeepSeek says product prices may vary, this page should be rechecked before using the numbers for production budgeting.
| Source | Used for |
|---|---|
| Official DeepSeek Models & Pricing | Current model names, API rates, context length, max output, and concurrency limits. |
| Official DeepSeek API Change Log | The V4.1-Flash release on September 10, 2026, the retirement of V4-Flash and Vision Exp, the V4-Pro continuation notice, the Pro-0813 GA rollout on August 13, and the pricing-transition date. |
| Official DeepSeek Vision Guide | Image formats, delivery methods, resizing, limits, and per-image token billing. |
| Official DeepSeek Files API | Uploading images and referencing them by file_id. |
| Official DeepSeek Responses API Guide | The Responses API contract, JSON Schema output, and tools. |
| Official DeepSeek V4 Release Notice | Historical alias mappings and the announced July 24, 2026 retirement cutoff. |
| Official DeepSeek Token & Token Usage | Token billing terminology and usage measurement. |
| Official DeepSeek Context Caching | Cache-hit rules, cache-miss behavior, and usage fields. |
| Official DeepSeek Rate Limit & Isolation | Concurrency limits and request behavior. |
| Official DeepSeek website | Free official DeepSeek web/app access and official product links. |
Pricing Update Log
| Date | Update |
|---|---|
| September 18, 2026 | Replaced the three-model catalog with the current two: deepseek-flash (DeepSeek-V4.1-Flash, released September 10, 2026, at $0.003 / $0.15 / $0.6 off-peak, with native image input up to 1,024 tokens per image) and deepseek-v4-pro (unchanged rates; API service continued after September 14, 2026). Recalculated the per-1K rates and worked examples, removed the retired Vision Exp rows, and linked the new V4.1-Flash guide. |
| August 21, 2026 | Added the experimental V4-Flash Vision Exp model (retired September 10, 2026), its Flash-matching rates at the time, image-token billing, Files and Responses support, the three-model comparison, and the shared cheapest-tier wording. |
| August 16, 2026 | Confirmed the peak/off-peak schedule is now active, moved current rates ahead of historical rows, and preserved all pre-cutover examples as dated calculations. |
| August 14, 2026 | Added Pro-0813 GA status, support for both models on Responses API, and the exact August 16 peak/off-peak pricing transition while preserving pre-cutover examples as dated calculations. |
| August 3, 2026 | Rechecked unchanged base rates, recorded Flash-0731 and Flash-only Responses API support, and added the then-announced peak-price policy without inventing an effective date. |
| July 29, 2026 | Rechecked the official DeepSeek Models & Pricing page, confirmed that the displayed V4 Flash and V4 Pro rates were unchanged, and synchronized the verification notes. |
| July 10, 2026 | Updated the page into a broader DeepSeek Pricing hub, removed expired V4 Pro promotion wording, verified official V4 Flash and V4 Pro API rates, added per-1K conversions, calculator, free chat comparison, improved FAQ, and schema. |
| April 26, 2026 | Previous page version focused mainly on DeepSeek API pricing, V4 models, cache-hit/cache-miss pricing, and legacy model names. |
DeepSeek Pricing FAQ
Is DeepSeek free?
As of September 3, 2026, DeepSeek’s official web chat and official app are presented as free to access. That is separate from the token-billed official API. Free access does not promise unlimited use, permanent terms, guaranteed availability, or an SLA.
How much does DeepSeek API cost?
DeepSeek API pricing is listed per one million tokens. deepseek-flash (DeepSeek-V4.1-Flash) is $0.003 cache-hit input / $0.15 cache-miss input / $0.6 output off-peak and $0.006 / $0.3 / $1.2 at peak. deepseek-v4-pro is $0.022 / $0.66 / $1.98 off-peak and $0.044 / $1.32 / $3.96 at peak. Images sent to V4.1-Flash add billed input tokens.
Is DeepSeek API free?
The public official pricing page does not document a permanent, universal free API tier. API usage is token-billed. A granted balance or promotional credit shown in an account is account-specific, so check its amount, eligibility, expiry date, and applicable models.
Does DeepSeek have a paid consumer subscription?
The public official sources checked on September 3, 2026 present the web/app chat experience as free and the API as token-billed; they do not show a ChatGPT Plus-style consumer subscription plan. Product plans can change, so verify the current official website and app-store listing.
Does DeepSeek charge per request or per token?
DeepSeek API billing is token-based, not fixed per request. Each request cost depends on cache-hit input tokens, cache-miss input tokens, image input tokens where applicable, and output tokens.
What is DeepSeek pricing per 1M tokens?
DeepSeek API pricing per 1M tokens is split by model, token type, and UTC billing window. Apply the current off-peak or peak row for the exact model and request time.
What is DeepSeek pricing per 1K tokens?
Divide the applicable official per-1M-token rate by 1,000. The table above shows the current off-peak and peak per-1K rates for deepseek-flash and deepseek-v4-pro.
What is the cheapest current DeepSeek API model?
deepseek-flash (DeepSeek-V4.1-Flash) has the lowest listed current token rate on every line, and it handles both text and images. Images add billed input tokens, so equal token rates do not mean equal total request cost.
Does image input cost extra on DeepSeek V4.1-Flash?
There is no separate image rate. Each image is converted into input tokens (up to 1,024 per image) and billed at the ordinary input rate for the window, so requests with images can still cost more overall than text-only requests.
How do V4.1-Flash and V4 Pro differ?
V4.1-Flash (deepseek-flash) is the lower-priced default, with native image input and a 2,500 concurrency limit. V4 Pro (deepseek-v4-pro) is text only, costs more on every rate line and has a 500 concurrency limit; test it on representative tasks and use it only when measured results justify the higher price.
What is cache-hit pricing?
Cache-hit pricing applies to input tokens that DeepSeek can serve from context caching because the relevant prefix is already persisted and matched. Cache-hit input tokens are much cheaper than cache-miss input tokens.
What is cache-miss pricing?
Cache-miss pricing applies to input tokens that require fresh processing. It is higher than cache-hit pricing, while output tokens are billed separately at the selected model’s output rate.
Are repeated prompts always billed at cache-hit pricing?
No. DeepSeek context caching works on a best-effort basis and depends on persisted matching prefixes. Track prompt_cache_hit_tokens and prompt_cache_miss_tokens in the API response to understand the real split for each request.
How can I reduce DeepSeek API costs?
Use V4.1-Flash for text and images at the lowest rate, disable thinking where reasoning is not needed, reserve Pro for tasks that show a measured benefit, reuse stable prefixes where useful, cap output length, minimize unnecessary image count or resolution, and monitor actual cache-hit, cache-miss, image-input, and output usage.
Are deepseek-chat and deepseek-reasoner still supported?
The legacy names are not among the two IDs in the current pricing matrix (deepseek-flash and deepseek-v4-pro). Although a bounded July 28 test observed them route to V4 Flash after returning HTTP 400 on July 25, that is unlisted and unstable compatibility behavior. Do not use the old aliases in new code, production fallbacks, or cost models.
Where can I verify official DeepSeek pricing?
Use the official DeepSeek Models & Pricing page. That page is the source of truth for current public API rates. Chat-Deep.ai is independent and does not control official DeepSeek API pricing.
Is Chat-Deep.ai affiliated with DeepSeek?
No. Chat-Deep.ai is independent and is not affiliated with, endorsed by, or operated by DeepSeek. It does not sell official DeepSeek API keys, credits, or billing plans.
For questions about this independent publication, our guides, tests, and privacy practices, see the Chat-Deep.ai site FAQ.
Recommended Next Pages
DeepSeek AI Explained — explore our independent guides and tests.
DeepSeek API Cost Calculator — estimate per-request, daily, monthly, and yearly costs.
DeepSeek API Guide — learn setup, API calls, model selection, and examples.
DeepSeek Context Caching — understand cache hits, cache misses, and prompt reuse.
DeepSeek Models — compare available DeepSeek models and usage patterns.
DeepSeek V4.1-Flash — specifications, pricing, native image input, thinking modes, and migration from retired IDs.