DeepSeek Subscription in 2026: Free App, No Paid Plan, API From $0.15 per 1M Tokens
Updated: 7 days ago

Updated: November 14, 2026
DeepSeek has no paid consumer subscription: the DeepSeek app is free, the web chat lists no price or plan, and only the developer API is paid, billed per token and per call from a prepaid balance. There is no Plus, Pro or Ultra tier to buy for the chat assistant, and DeepSeek's download page describes the app as a way to "chat with leading AI models for free."
The paid side is the API. As listed on DeepSeek's official pricing page on November 14, 2026, the default model deepseek-flash (DeepSeek-V4.1-Flash-1017) costs $0.15 per 1M input tokens off-peak and $0.30 at peak, with output at $0.60 and $1.20 per 1M tokens and cached input from $0.003. On November 13, 2026 DeepSeek also opened its application APIs to developers, so web search, web fetch, file parsing with OCR, speech to text and text to speech now have their own price lines. The table below sums up what is free and what is paid.
........
DeepSeek access route | What you pay | Model or service behind it | Best fit |
DeepSeek app (iOS, Android) | Free, no subscription | DeepSeek-V4.1-Flash-1017 | Everyday chat, search, files |
DeepSeek web chat | No price or plan listed | DeepSeek-V4.1-Flash-1017 | Desktop browser use |
DeepSeek API, deepseek-flash | From $0.15 per 1M input tokens | DeepSeek-V4.1-Flash-1017 | Apps, agents, batch jobs |
DeepSeek API, deepseek-v4-pro | From $0.66 per 1M input tokens | DeepSeek-V4-Pro-0813 | Teams that stayed on V4 Pro |
Search and Fetch APIs | $0.75 per 1,000 searches or fetches | DeepSeek's own search engine | Web grounding for agents |
File parsing and OCR API | $0.225 per 1,000 pages | DeepSeek document parser | PDFs, scans, document pipelines |
Speech APIs (ASR, TTS) | From $0.058 per hour of audio | DeepSeek speech models | Voice apps, transcription |
Self-hosting open weights | Free MIT license, own hardware | DeepSeek-V4.1-Flash-1017 | Large teams with GPUs |
........
DeepSeek publishes no message caps for the free app, so the practical limit for consumer users is server availability during busy hours. Developers get a precise price list instead, and the sections below explain each part of it with cost examples computed for real workloads.
··········
IS DEEPSEEK FREE? THE DEEPSEEK APP COSTS NOTHING AND HAS NO PAID TIER.
The consumer DeepSeek assistant is free on iOS and Android, and the browser version linked from DeepSeek's home page as DeepSeek Web shows no price or plan on DeepSeek's official pages. DeepSeek's download page offers the iOS app through the App Store and the Android app as a direct APK download, and it mentions no subscription, paid tier or in-app plan anywhere.
DeepSeek's official pages list no monthly consumer plan in November 2026, so a page or app that asks for a "DeepSeek Premium" payment is not an official DeepSeek product. Third-party apps with similar names exist in the app stores, which makes the developer name the simplest check before installing.
The model behind the free app changes over time. DeepSeek's API change log for October 17, 2026 states that the web and app models were updated to DeepSeek-V4.1-Flash-1017, the same model served through the API as deepseek-flash. Free users therefore get the current Flash model, with a DeepThink toggle for the reasoning mode and a Search toggle for web results, both shown on DeepSeek's home page.
DeepSeek does not publish a daily message limit, file quota or usage cap for the free chat. When demand is high the service can slow down or show a busy message, and DeepSeek offers no paid way to skip that queue.
··········
DEEPSEEK API PRICING IN NOVEMBER 2026: DEEPSEEK-FLASH AND DEEPSEEK-V4-PRO.
The DeepSeek API is pay-as-you-go. Developers top up a balance on the DeepSeek platform and every request is charged as tokens multiplied by price, with no monthly seat or minimum plan. When an account holds both a granted balance and a topped-up balance, DeepSeek deducts from the granted balance first, according to its pricing page.
Two models are on the price list. deepseek-flash serves DeepSeek-V4.1-Flash-1017, released as an update on October 17, 2026 with unchanged pricing. deepseek-v4-pro serves DeepSeek-V4-Pro-0813, which DeepSeek kept on sale after September 14, 2026 with billing unchanged, according to its change log, after first announcing that V4 Pro traffic would be routed to V4.1-Flash.
........
Price per 1M tokens (USD) | deepseek-flash off-peak | deepseek-flash peak | deepseek-v4-pro off-peak | deepseek-v4-pro peak |
Input, cache hit | $0.003 | $0.006 | $0.022 | $0.044 |
Input, cache miss | $0.15 | $0.30 | $0.66 | $1.32 |
Output | $0.60 | $1.20 | $1.98 | $3.96 |
........
Prices are as listed on the official DeepSeek pricing page on November 14, 2026. Off-peak rates are exactly half of the peak rates for both models, and DeepSeek states on the same page that it may adjust prices, so production budgets need a periodic check against the live page.
Both models offer a 1M-token context length and up to 384K output tokens, with thinking mode on by default and a non-thinking mode available. Both support JSON output, tool calls, the Responses API, the Anthropic-compatible endpoint and chat prefix completion in beta, while FIM completion works only in non-thinking mode. Vision input is supported on deepseek-flash and not on deepseek-v4-pro, and the pricing page lists a concurrency limit of 2,500 for Flash and 500 for V4 Pro.
··········
DEEPSEEK PEAK AND OFF-PEAK HOURS: WHEN THE API COSTS HALF.
Peak pricing applies from 01:00 to 04:00 UTC and from 06:00 to 10:00 UTC, Monday through Friday. Weekends are off-peak all day, and so are Chinese public holidays, which DeepSeek excludes from peak time entirely. For a US team on Eastern Standard Time, the two peak windows run from 8:00 to 11:00 p.m. and from 1:00 to 5:00 a.m., so daytime traffic in the Americas already runs at off-peak rates.
DeepSeek introduced this two-rate system with a change that took effect at 16:00 UTC on August 16, 2026, and kept it when V4.1-Flash launched with lower prices on September 10, 2026. Schedulable work such as indexing, evaluation runs and document batches can be moved outside the peak windows to cut the token bill in half without changing a single prompt. The application APIs described in the next section have a single price at all hours.
··········
DEEPSEEK WEB SEARCH, OCR AND VOICE API PRICES SINCE NOVEMBER 13, 2026.
DeepSeek's change log for November 13, 2026 adds application APIs for deepseek-flash across the Messages, Responses and Chat Completions APIs, covering web search, file input, URL reading and model safety, plus standalone Search and Fetch, OCR, speech to text (ASR) and text to speech (TTS) endpoints. DeepSeek presents the release as the full DeepSeek app experience offered through the API, and the details and launch context are in our report on DeepSeek's new web search, OCR and voice APIs.
........
DeepSeek application API | Unit | List price | Current price |
Search (text, image) | Per 1,000 searches | $0.75 | $0.75 |
Web fetch | Per 1,000 fetches | $0.75 | $0.75 |
File parsing and OCR | Per 1,000 pages | $0.225 | $0.225 |
Text to speech (TTS) | Per 10,000 characters | $0.45 | $0.225 (limited-time, 50% off) |
Speech to text (ASR) | Per hour of audio | $0.116 | $0.058 (limited-time, 50% off) |
........
The Search and Fetch guide in DeepSeek's documentation says that a request is billable only when it returns HTTP 200 and produces a result, while failed requests, empty search results and timeouts are not billed. The number of results requested does not change the price, and the model-driven web tools inside a chat request are billed at the same search and fetch prices. Calls triggered from coding tools such as Claude Code or Codex are also billed per successful request, according to the same page. The pricing page gives no end date for the 50% discount on the speech APIs.
Each service has its own limits. The Search API allows 100 requests per second per account and the Fetch API 25, with a further per-domain limit on fetches, and both limits are separate from model concurrency. The TTS endpoint accepts up to 10,000 billed characters per request, returns at most 30 minutes of audio, offers two voices and counts each Chinese character as two characters for billing.
··········
[[ADPLUS_300x250_1]]
WHAT DEEPSEEK COSTS FOR A REAL WORKLOAD: FLASH VS V4 PRO AND THE NEW APIS.
To turn per-token prices into a monthly figure, Data Studios computed a customer-support chatbot handling 1,000 conversations a day for 30 days. Each request sends 3,000 input tokens, of which 2,000 are a stable system prompt served from cache and 1,000 are new text, and returns 500 output tokens. A second scenario adds a monthly batch job of 10M uncached input tokens and 2M output tokens. All prices come from the official tables above.
........
Monthly scenario | deepseek-flash | deepseek-v4-pro |
Chatbot, 30,000 requests, all at peak rates | $27.36 | $101.64 |
Chatbot, 30,000 requests, all off-peak | $13.68 | $50.82 |
Batch job: 10M uncached input + 2M output, peak | $5.40 | $21.12 |
Same batch job, off-peak | $2.70 | $10.56 |
Same batch job, fully cached input, off-peak | $1.23 | $4.18 |
........
The arithmetic is tokens multiplied by the listed price per million. For the chatbot on Flash at peak, 60 million cached tokens cost $0.36, 30 million uncached tokens cost $9.00 and 15 million output tokens cost $18.00, for $27.36 in total. Output dominates the bill, so trimming verbose answers saves more than any other change once the system prompt is cached. V4 Pro costs about 3.7 times as much as Flash on the chatbot scenario, and DeepSeek's September release note says that tests by multiple parties put V4.1-Flash ahead of V4-Pro on performance, cost and speed.
The application APIs add a separate line to the bill. A research assistant that makes 10,000 successful searches and 5,000 page fetches a month, parses 20,000 document pages, transcribes 100 hours of audio and reads out 1M characters would pay $7.50 for search, $3.75 for fetch, $4.50 for OCR, $5.80 for speech to text and $22.50 for text to speech at the current prices, or $44.05 a month before token costs. At the list prices for speech, the same workload rises to $72.35, because transcription would cost $11.60 and speech synthesis $45.00. Developers comparing DeepSeek with other providers can find official prices side by side in our comparison of DeepSeek V4.1 Flash, Claude Opus 5 and GPT-5.6 Sol.
··········
OLD DEEPSEEK MODEL NAMES: DEEPSEEK-CHAT, DEEPSEEK-REASONER AND V4 FLASH.
Many tutorials still refer to deepseek-chat and deepseek-reasoner, the API names used in the V3 and R1 era. DeepSeek's change log for April 24, 2026 scheduled both names for discontinuation on July 24, 2026, and until then they pointed to V4-Flash in non-thinking and thinking mode. Neither name appears on the current pricing page, so integrations built on them need the model field changed to deepseek-flash.
........
Model name in old code | Status in November 2026 | What to use now |
deepseek-chat | Scheduled for discontinuation on July 24, 2026 | deepseek-flash, non-thinking mode |
deepseek-reasoner | Scheduled for discontinuation on July 24, 2026 | deepseek-flash, thinking mode |
deepseek-v4-flash | Model retired, served by V4.1-Flash-1017 at Flash price | deepseek-flash |
deepseek-v4-flash-vision-exp | Model retired, served by V4.1-Flash-1017 at Flash price | deepseek-flash |
deepseek-v4-pro | Still served by DeepSeek-V4-Pro-0813 at its own rates | Keep, or test deepseek-flash |
........
The retired V4 Flash names still answer requests, and DeepSeek's September change log called that routing temporary. Switching to deepseek-flash removes the risk of a sudden break when the aliases are switched off. Readers who need the full developer billing picture, including balance top-ups and granted credit, can continue with our guide to DeepSeek API pricing and developer access.
··········
SELF-HOSTING DEEPSEEK: FREE WEIGHTS, PAID HARDWARE.
DeepSeek-V4.1-Flash is an open-weight model. Its Hugging Face page states that "this repository and the model weights are licensed under the MIT License," which allows commercial use, modification and private deployment without paying DeepSeek. The October update, DeepSeek-V4.1-Flash-1017, is published on Hugging Face under the same license and is described there as a minor update to DeepSeek-V4.1-Flash.
The weights are free, but running them is costly. Hugging Face lists the checkpoint at 763B parameters, and the DeepSeek-V4.1-Flash model card describes 552B backbone parameters with 8B parameters activated per token during prefill and 16B during decoding, plus support for contexts of up to one million tokens. Serving a model of this size needs multi-GPU accelerator capacity, so self-hosting suits organizations with steady high volume, strict data-location rules or an existing inference cluster. For occasional use, the free app or the pay-as-you-go API is far cheaper than buying or renting the hardware.
Self-hosting also changes who handles data. With the official app and API, prompts are processed on DeepSeek's servers under its terms and privacy policy, while a self-hosted deployment keeps prompts inside the operator's own infrastructure. Technical details of the latest release are covered in our report on DeepSeek V4.1 Flash 1017 and its open weights.
··········
HOW TO CHECK WHAT YOU PAY FOR DEEPSEEK.
Consumer users have little to check: the official app is free, and DeepSeek offers no subscription to cancel. A payment request from an app named "DeepSeek" usually means a third-party clone, so the developer name in the store listing is the first thing to verify.
API users can follow five steps to keep costs predictable. First, set the model field to deepseek-flash unless a tested reason exists to stay on deepseek-v4-pro. Second, keep the long, stable part of each prompt identical between calls so that it is billed at the cache-hit rate. Third, schedule batch jobs outside the weekday windows of 01:00-04:00 and 06:00-10:00 UTC. Fourth, count search, fetch and OCR calls separately from tokens, since each successful call has its own price. Fifth, budget speech features at the list price as well as the discounted one, because DeepSeek has not said when the 50% discount ends.
The official pricing page remains the reference for any later change, since DeepSeek reserves the right to adjust prices. DeepSeek has not announced a date for a V4.1 Pro model or for the shutdown of the remaining legacy aliases.
··········
DEEPSEEK SUBSCRIPTION IN 2026: FREE CHAT, PAID API, OPEN WEIGHTS.
DeepSeek still has no consumer subscription in November 2026. The app on iOS and Android is free and the web chat lists no price; both run the current DeepSeek-V4.1-Flash-1017 model, with no published usage caps and no paid upgrade.
Money changes hands only on the API, where deepseek-flash costs $0.15 per 1M uncached input tokens and $0.60 per 1M output tokens off-peak, twice that at peak, and deepseek-v4-pro remains available at roughly three to four times the Flash rates. Since November 13 the application APIs add per-call prices of $0.75 per 1,000 searches or fetches, $0.225 per 1,000 OCR pages and discounted speech rates of $0.058 per audio hour and $0.225 per 10,000 characters. Self-hosting through the MIT-licensed weights removes the API bill and replaces it with infrastructure cost, which makes it a choice for large, steady workloads rather than a free alternative for individual users.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org




