DeepSeek

Free consumer chat, with separate APIs and model weights

by DeepSeek

Overview

DeepSeek provides free consumer chat at chat.deepseek.com and in its app, alongside a separately billed developer platform and downloadable model weights. Its September 2026 iOS release notes say Instant, Expert and Vision have been unified. We did not log in to verify the current web controls, so an older DeepThink button layout is not presented as universal.

An official January 2025 app announcement documents search, file upload and text extraction. It does not establish today’s consumer file-count, size, page or usage limits. This review has not verified a ten-PDF research workflow, original-page citations or Word/PDF report export. A free entry point is not unlimited processing or proof of a complete research deliverable.

The deepseek-flash API is separately metered, with peak/off-peak token rates. Its Files API accepts JPEG, PNG, GIF and WebP images, up to 64 MiB each; it is not a direct PDF-upload API. The V4.1-Flash model card documents image/text input, text output, a million-token context and MIT-licensed weights; these are model specifications, not a promise about free chat. Self-hosting requires inference resources and your own file/search workflow.

The consumer privacy policy permits model-improvement use with a training opt-out right, describes storage/processing in China and says search keywords may be passed to third parties. Check those conditions against your organisation’s requirements. We have not tested this product’s output quality, file coverage or account-specific limits.

Commercial terms are stated rather than implied: the Terms of Use assign DeepSeek's rights in the Outputs to you and allow a wide range of uses, including personal use, academic research, developing derivative products and training other models such as distillation, provided the use is lawful and follows the terms. The same terms forbid using DeepSeek's trademarks, logos or brand features without permission, and they do not split the commercial conditions between the free and paid consumer tiers (DeepSeek Terms of Use, read 12 October 2026).

Editor's take

A candidate for free consumer chat or a separately budgeted developer integration. Choose the actual channel first: consumer file/export limits remain unverified, and model context length does not establish a ten-PDF deliverable. No price/performance ranking or unlimited-use claim is supported by this review.

Our reading of who this suits, based on the sources above. Opinion, offered to help you decide - not a lab result.

Key Features

Free consumer web/app chat; API usage is billed separately
Search and file text extraction documented in the January 2025 app announcement
September 2026 iOS notes unify Instant, Expert and Vision; current web controls were not checked
deepseek-flash API with separate cache-hit, cache-miss and output rates
V4.1-Flash image/text-to-text model with MIT-licensed weights for self-hosting
Image-only Files API: JPEG/PNG/GIF/WebP, 64 MiB per file and 25 GiB storage per user

Use Cases

Consumer questions and file-text tasks after checking the current account limits
An API integration budgeted using deepseek-flash rates and the actual input/output mix
Self-hosting the specified open-weight model with a separately built document and retrieval pipeline

Pros

✓Consumer chat has an official free-access listing
✓The API publishes rate categories that can be budgeted for the specified model
✓The specified V4.1-Flash release has an explicit MIT licence and downloadable weights

Cons

✕Current consumer PDF count, file size, page handling and report-export formats were not verified
✕Files API image support does not establish direct PDF support; files without an expiry can remain stored indefinitely
✕The September 10 announcement and current price table disagree about the deepseek-v4-pro route
✕Consumer data may be used for model improvement; training opt-out, China processing and third-party search transfer need consideration

Frequently Asked Questions

Is DeepSeek free, and does that include the API?

Consumer chat is listed as free. The API deducts metered usage from a prepaid balance; it is not a monthly or annual chat subscription. Free chat does not establish unlimited usage, and this review did not verify current consumer file quotas.

What does deepseek-flash cost?

USD per million tokens: off-peak cache-hit input $0.003, cache-miss input $0.15 and output $0.60; peak rates are $0.006, $0.30 and $1.20 respectively. Peak windows are Monday–Friday 01:00–04:00 and 06:00–10:00 UTC, excluding Chinese public holidays. The API table lists 1M context and a maximum 384K output. These are API terms, not free-chat allowances.

Why is no deepseek-v4-pro price recommended here?

The September 10 announcement says this route switches to V4.1-Flash and Flash billing from September 14 at 04:00 UTC, while the price table still lists V4-Pro-0813 with different rates and no vision. This unresolved official discrepancy should be checked before estimating that route; the explicit deepseek-flash rates are kept separate.

Can I upload PDFs through the Files API?

The reviewed Files API specifies JPEG, PNG, GIF and WebP images, a 64 MiB file limit and 25 GiB user storage. Direct PDF upload is not established by that interface; an API document workflow needs separate extraction or page processing. Files without an expiry may be retained indefinitely.

Can ten public PDFs become a cited Chinese comparison within US$20?

The free consumer price condition is documented, but ten-file capacity, complete coverage, traceable PDF-page citations, disagreements and report export were not verified. It is not a qualified complete-task recommendation. Low API token prices also do not supply a finished research interface.

Do open weights or privacy controls make all use local?

No. The V4.1-Flash weights have an MIT licence, but self-hosting needs your own inference and file/search setup. Hosted consumer processing is separate: the policy describes model-improvement use, a training opt-out right, China storage/processing and third-party search-keyword transfer. Copying or sharing a chat is not proof of cited Word/PDF export.

Task guides for this product

T11 · Which assistants can be used from mainland China

Key facts and where they come from

PriceNot established: The US App Store listing marks the official DeepSeek AI Assistant for iPhone and iPad as Free. The selected consumer web channel’s regional purchase price and eligibility remain unestablished; that storefront listing does not establish worldwide free or unlimited service, and API billing is separate.Source: DeepSeek consumer chat · DeepSeek official iOS listing and release notes
AvailabilityNot established: Country-by-country distribution and actual reachability of the selected product and plan are not established by the registered sources. A platform list, hosting location or payment-region note does not establish access from a particular country.Source: DeepSeek product entry points · DeepSeek consumer chat
Commercial useDocumented: Read on 12 October 2026. Assigns rights in the Outputs to the user and permits wide use including training other models, requires lawful use and compliance with the terms, and forbids unlicensed use of DeepSeek trademarks. The terms are not split between free and paid consumer tiers.Source: DeepSeek - Terms of Use
Training on your dataDocumented: Historical scope read on 2026-10-12, not rechecked in full on 2026-10-15: February 10, 2026 policy covers input/files/history processing, model improvement, training opt-out, China storage/processing and third-party search-keyword transfer. Chat copy/share features do not establish cited document export.Source: DeepSeek consumer privacy policy
API billingDocumented: The API bills per million model input and output tokens with peak and off-peak rates; those rates do not establish consumer document capacity or a workload within CNY100. Historical scope read on 2026-10-12, not rechecked in full on 2026-10-15: deepseek-flash USD/million tokens: off-peak hit/miss/output 0.003/0.15/0.60; peak 0.006/0.30/1.20. Peak weekdays 01:00–04:00 and 06:00–10:00 UTC exclude Chinese public holidays. API context 1M, max output 384K. The Pro table conflicts with the September routing announcement.Source: DeepSeek API pricing
Export formatsNot established: Current consumer Word/PDF or other report-download formats are not established; the app announcement and privacy record do not verify cited-document export.Source: DeepSeek app announcement, January 2025 · DeepSeek consumer privacy policy
Usage limitsDocumented: JPEG/PNG/GIF/WebP, up to 64 MiB/file and 25 GiB storage/user; no direct PDF support established here, and files without an expiry can be retained indefinitely.Source: DeepSeek image Files API

Official sources

The vendor pages we read for the claims on this page, what each one states, and the day we read it.

DeepSeek product entry points ↗
Separates Web, App, API Platform and Harness; these are different access channels.
Read on September 23, 2026
DeepSeek consumer chat ↗
Consumer chat entry; this review did not log in or verify its current controls and file limits.
Read on September 23, 2026
DeepSeek official iOS listing and release notes ↗
Lists free access; September 11–12, 2026 releases 2.5.0/2.5.1 describe unifying Instant, Expert and Vision. This does not verify every web control.
Read on September 23, 2026
DeepSeek app announcement, January 2025 ↗
Documents search, thinking, file upload/text extraction and history sync; it does not establish current consumer PDF quotas or report exports.
Read on September 23, 2026
V4.1-Flash official model card ↗
Image/text input, text output, million-token model context and MIT-licensed repository/weights; self-hosting needs infrastructure and workflow integration.
Read on September 23, 2026
DeepSeek API pricing ↗
Historical registered scope from 2026-10-12; the entire statement was not re-read on 2026-10-15: deepseek-flash USD/million tokens: off-peak hit/miss/output 0.003/0.15/0.60; peak 0.006/0.30/1.20. Peak weekdays 01:00–04:00 and 06:00–10:00 UTC exclude Chinese public holidays. API context 1M, max output 384K. The Pro table conflicts with the September routing announcement. Official body read on 2026-10-15, limited to this scope: API documentation bills per million model input/output tokens, lists 1M context and maximum384K output, with peak/off-peak rates. API token prices do not establish consumer document-reading limits or thatCNY100 covers a workload. The API bills per million model input and output tokens with peak and off-peak rates; those rates do not establish consumer document capacity or a workload within CNY100. Historical scope read on 2026-10-12, not rechecked in full on 2026-10-15: deepseek-flash USD/million tokens: off-peak hit/miss/output 0.003/0.15/0.60; peak 0.006/0.30/1.20. Peak weekdays 01:00–04:00 and 06:00–10:00 UTC exclude Chinese public holidays. API context 1M, max output 384K. The Pro table conflicts with the September routing announcement.
Read on October 15, 2026
V4.1-Flash routing announcement ↗
September 10 announcement says deepseek-v4-pro routes to V4.1-Flash and Flash billing from September 14 at 04:00 UTC; the reviewed price table still differs.
Read on September 23, 2026
DeepSeek image Files API ↗
JPEG/PNG/GIF/WebP, up to 64 MiB/file and 25 GiB storage/user; no direct PDF support established here, and files without an expiry can be retained indefinitely.
Read on September 23, 2026
DeepSeek consumer privacy policy ↗
Historical registered scope from 2026-10-12; the entire statement was not re-read on 2026-10-15: February 10, 2026 policy covers input/files/history processing, model improvement, training opt-out, China storage/processing and third-party search-keyword transfer. Chat copy/share features do not establish cited document export. Official body read on 2026-10-15, limited to this scope: The policy states direct collection, processing and storage of personal data in thePRC and appoints a privacy representative forEU/UK. This does not establish actual sign-up countries. Historical scope read on 2026-10-12, not rechecked in full on 2026-10-15: February 10, 2026 policy covers input/files/history processing, model improvement, training opt-out, China storage/processing and third-party search-keyword transfer. Chat copy/share features do not establish cited document export.
Read on October 15, 2026
DeepSeek - Terms of Use ↗
Read on 12 October 2026. Assigns rights in the Outputs to the user and permits wide use including training other models, requires lawful use and compliance with the terms, and forbids unlicensed use of DeepSeek trademarks. The terms are not split between free and paid consumer tiers.
Read on October 12, 2026

Compare with other tools in Chat Assistants

Product	Maker	Pricing	Launched	Shared tags
Grok	xAI	Freemium	2023-11-03	Chat, Reasoning
Mistral Vibe	Mistral AI	Freemium	2024-02-26	Chat
Kimi	Moonshot AI	Freemium	2023-11-16	Chat
Doubao	ByteDance	Freemium	2023-08	Chat
Qwen Studio	Alibaba	Free	2023-04	Chat
Gemini	Google	Freemium	2023-03-21	Chat