Your privacy, your call. No advertising cookies. May we count your visit in aggregate (Cloudflare, no cookies) and remember where you came from? If you reject, we won’t measure your visit. · Cookies · Privacy

Privacy settings

Choose what we measure and what we keep in this browser. We never sell or share your data.

Essential

Remember your theme, language, saved items and this choice. Without them the site cannot work the way you ask.

Always on

Keeps the source of your first and latest visit for 90 days (utm parameters and referring site). It only leaves your browser if you subscribe to the newsletter.

Cloudflare Web Analytics counts visits in aggregate: page, referring site, country, browser and load times. No cookies, no identifiers and no profiling; nothing is stored in your browser. It only loads if you accept it. If you reject it or switch it off, it does not load (or stops loading from the next page) and we do not count your Telegram clicks either.

Cookies · Privacy
EN
My account
Comparison · AI models

DeepSeek vs ChatGPT: privacy, prices and API (2026)

DeepSeek or ChatGPT: where your data is stored, what each costs, API prices and how to use DeepSeek without sending data to China. With official sources.

By Cognenza editorial team · Updated · Euro prices as published for Spain; they may differ in your country.

Verdict

For everyday use with personal data, ChatGPT with training turned off. DeepSeek is worth it for its API prices and MIT weights, not for privacy: its app stores data in China.

Main differences

  • Data: DeepSeek stores it in China; OpenAI lets you turn off training in Data controls, even without logging in.
  • Weights: DeepSeek publishes MIT-licensed weights that you can host in your own cloud; GPT-6 is closed.
  • API: deepseek-flash costs USD 0.30 per million input tokens at peak time (USD 0.15 off-peak); gpt-6.1-sol, USD 2.00.
  • Plans: in Spain ChatGPT has Free, Go (€8 per month), Plus (€23 per month) and Pro (from €103 per month).

Side by side

DeepSeek vs ChatGPT: privacy, prices and API (2026)
DeepSeek-V4.1-Flash-1017 (and DeepSeek-V4-Pro)GPT-6 (Luna, Sol, Astra) and GPT-6.1 Sol
OrganisationDeepSeek · ChinaOpenAI · United States
Release17 October 20267 October 2026
Open weightsYes · MITNo · proprietary
Parametersnot officially publishednot officially published
Context1,000,000 tokens1,050,000 tokens
Inputtext, imagetext, image, voice
LocalYesNo
FreeFree downloadable weightsFree plan
Paid plan fromnot officially publishedGo: €8 per month
API fromUSD 0.15 per million input tokensUSD 0.10 per million input tokens
Trains on your data?Yes (the policy includes training and improving models)Yes on consumer plans unless you turn it off; Business, Enterprise and Edu do not train by default
How to avoid itRight to object by email to privacy@deepseek.comSettings > Data controls > “Improve the model for everyone” (also works without logging in; saved per browser). Temporary chats are not used for training and are kept for up to 30 days. Thumbs-up/down feedback may be used for training
Where data is storedPeople’s Republic of Chinanot officially published
GDPRThe DPF does not apply (Chinese company); the policy recognises the right to complain to your local data protection authority (in the UK, the ICO)not officially published

API rates in US dollars (USD) per million tokens, as published by each company. Benchmarks are the manufacturer’s own figures: they are not Cognenza tests and cannot be compared with each other.

Frequently asked questions

Is DeepSeek free?

The web chat and the app are free to use; the API is pay-as-you-go. Whether there are paid consumer plans is still to be confirmed.

Is DeepSeek safe to use?

Its policy says it stores data in China and may use it for training. You can object by writing to privacy@deepseek.com and, according to its policy, complain to your local data protection authority (in the UK, the ICO). Do not paste sensitive data into it.

Can I use DeepSeek without sending data to China?

Yes, by running its open weights on your own infrastructure or through a provider that hosts it in Europe.

Which is better for coding?

Pending our own tests. The published figures come from the manufacturers themselves and cannot be compared with each other.

Model by model

DeepSeek-V4.1-Flash-1017 (and DeepSeek-V4-Pro)

Updated

Organisation
DeepSeek · China
Release
17 October 2026
Open weights
Yes · MIT · huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017
Parameters
not officially published
Context
1,000,000 tokens
Input
text, image
Access
Local use
Yes · Minimum memory: not officially published · Tools: ollama, lm-studio
Consumer plans
not officially published
Trains on your data?
Yes (the policy includes training and improving models)
How to avoid it
Right to object by email to privacy@deepseek.com
Where data is stored
People’s Republic of China
GDPR
The DPF does not apply (Chinese company); the policy recognises the right to complain to your local data protection authority (in the UK, the ICO)
Privacy policy
cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html

API rates (USD per million tokens)

API rates (USD per million tokens): DeepSeek-V4.1-Flash-1017 (and DeepSeek-V4-Pro)
ModelInputCached inputOutput
deepseek-flashUSD 0.30 (peak) · USD 0.15 (off-peak)USD 0.006 (peak) · USD 0.003 (off-peak)USD 1.20 (peak) · USD 0.60 (off-peak)
deepseek-v4-proUSD 1.32 (peak) · USD 0.66 (off-peak)USD 0.044 (peak) · USD 0.022 (off-peak)USD 3.96 (peak) · USD 1.98 (off-peak)

Off-peak rate = half the peak rate. Peak: 01:00–04:00 and 06:00–10:00 UTC, Monday to Friday (03:00–06:00 and 08:00–12:00 Madrid time in summer). api-docs.deepseek.com/quick_start/pricing · API rates in USD, 22 October 2026.

Notes

  • Web chat and app are free; whether there are paid plans is still to be confirmed.

Benchmarks

Manufacturer’s figure, not comparable across models.

Our own tests

Cognenza has not run its own tests on this model yet.

Sources

GPT-6 (Luna, Sol, Astra) and GPT-6.1 Sol

Updated

Organisation
OpenAI · United States
Release
7 October 2026
Open weights
No · proprietary · openai.com/policies/services-agreement
Parameters
not officially published
Context
1,050,000 tokens
Input
text, image, voice
Access
Local use
No
Consumer plans
  • Free: Free
  • Go: €8 per month
  • Plus: €23 per month
  • Pro: from €103 per month
Trains on your data?
Yes on consumer plans unless you turn it off; Business, Enterprise and Edu do not train by default
How to avoid it
Settings > Data controls > “Improve the model for everyone” (also works without logging in; saved per browser). Temporary chats are not used for training and are kept for up to 30 days. Thumbs-up/down feedback may be used for training
Where data is stored
not officially published
GDPR
not officially published
Privacy policy
help.openai.com/en/articles/7730893-data-controls-faq

API rates (USD per million tokens)

API rates (USD per million tokens): GPT-6 (Luna, Sol, Astra) and GPT-6.1 Sol
ModelInputCached inputOutput
gpt-6.1-sol (Above 272K input tokens: input ×2 and output ×1.5. US or EU data residency: +10%)USD 2.00USD 0.10USD 10.00
gpt-6-lunaUSD 0.10—USD 0.50
gpt-6-solUSD 2.00—USD 10.00
gpt-6-astra (Above 272K input tokens: input and cache ×2, output ×1.5 (whole request). Batch and Flex: 50%. Fast: ×2. Regional data residency: +10%)USD 10.00USD 1.00USD 50.00

developers.openai.com/api/docs/pricing · API rates in USD, 22 October 2026.

Notes

  • Free and Go use GPT-6 Luna; Plus, Pro, Business and Enterprise use GPT-6 Sol. GPT-6.1 Sol is in ChatGPT Work and Codex from Plus upwards.

Our own tests

Cognenza has not run its own tests on this model yet.

Sources

Every figure comes from the company’s official page and shows the date we checked it. If something is not officially published, we say so; we never estimate it. How we work.

Weekly AI newsletter

Every Tuesday at 07:00 UK time: the week’s AI news with sources, a guide and a tool. Free, no ads, no ad tracking. Unsubscribe in one click. What is inside and a sample issue

Controller: as stated in the privacy policy. Purpose: sending you the weekly newsletter (on Tuesdays) and, when you sign up, 2 welcome emails with resources (days 3 and 7 after confirming), and measuring, in aggregate, which pages and campaigns bring sign-ups. Minimum age: 16. Legal basis: your consent. We store your email, language, dates and source data (no IP). Recipients: Google (spreadsheet and Gmail), Resend (email delivery) and Cloudflare (hosting and bot check) as processors; no data is shared. Retention: until you unsubscribe; unconfirmed sign-ups are deleted after 7 days. Rights: access, rectification, erasure, objection, restriction and portability at contacto@cognenza.com; complaint to the data protection authority of your country of residence (such as the ICO in the UK or the DPC in Ireland) or to the Spanish AEPD as lead authority. Privacy policy.