ChatGPT vs DeepSeek: Is the Free Challenger Actually Better?

Quick pick — the answer in 30 seconds

Your job	Pick	Why, in one line
Everyday non-sensitive text, coding help, study, long documents	DeepSeek free	$0, near-Plus-level text ability, if China storage and “server is busy” are acceptable
Mixed work: images, voice, research reports, memory, projects, apps	ChatGPT Plus, $20/month	The only one of the two with image generation, voice, Deep Research, memory, Projects and Custom GPTs
Heavy chat + coding inside ChatGPT’s world (Codex, Work)	ChatGPT Plus $20 first, Pro $100 only when Plus caps bite	Plus is the cheapest tier with expanded Codex
Heavy cited research, not just web search	ChatGPT Plus $20	DeepSeek has web search only, no Deep Research agent
Developer / bulk API work, non-sensitive	DeepSeek API, pay-per-token	Flash off-peak $0.15 in / $0.60 out per million tokens, vs OpenAI Sol $2 / $10
Sensitive, regulated, client or government-contract work	Neither free app	Use ChatGPT Business/API, or self-hosted DeepSeek weights

The short answer

DeepSeek arrived with frontier-level benchmarks and a price tag of zero, and every ChatGPT user asked the same question: why am I paying? The honest answer depends on what you actually do with AI all day — and on a question most comparisons bury: where does your data go?

If your work is text — writing, coding, analysis, long documents — and it is not sensitive, DeepSeek free is genuinely hard to beat. It costs nothing, it handles long documents with a 1M-token API context, and on scoped coding and reasoning benchmarks it trades blows with ChatGPT’s Plus-level models.

If your work is mixed — images, voice, cited research reports, memory across chats, projects, apps, a desktop workflow — ChatGPT Plus is the only one of the two that actually has those tools. DeepSeek does not have image generation, voice conversation, a Deep Research agent, persistent memory, Projects or Custom GPTs. That gap, not raw model scores, is what justifies $20 for most mixed-work buyers.

And if your data is sensitive, regulated, under NDA, or for a government contractor, stop at price: hosted DeepSeek stores data in China, and consumer ChatGPT trains on chats by default unless you opt out. Neither free app is the right tool for that job. Use ChatGPT Business/API, or self-hosted DeepSeek weights — routing, not model choice, decides privacy.

Pricing — snapshot, 31 October 2026

Rewritten and re-checked 31 October 2026. Originally published 30 October 2026. Prices below are a snapshot checked 31 October 2026 on the vendors’ own pricing pages — those pages will change, so check at checkout for the final price.

	ChatGPT	DeepSeek
Free chat	$0. Unlimited text chats with GPT-6, subject to guardrails; uploads, image, voice, deep research, memory, Codex and Work are “limited”	$0. No subscription found on vendor pages. Any site selling a “DeepSeek Plus/Pro subscription” is a third party, not DeepSeek
Next step up	Go: $8/month (US; varies by country)	No paid chat tier
Flagship individual	Plus: $20/month, monthly only (no annual discount; $240/year arithmetic)	—
Top individual	Pro: from $100/month, tiers $100 / $200 / $500. $200 availability at checkout should be verified — trackers say new sign-ups to $200 paused since 10 September 2026	—
Business	Standard $20/seat/month billed annually ($25 monthly); Premium $100 annual ($125 monthly); Enterprise custom	—
API (separate bill from ChatGPT)	GPT-6 Luna $0.10 in / $0.50 out per million tokens; Terra $1 / $5; Sol $2 / $10; Astra $10 / $50	Flash off-peak $0.15 in (cache $0.003) / $0.60 out; peak double ($0.30 / $1.20). Pro off-peak $0.66 / $1.98; peak double. Peak = 01:00–04:00 and 06:00–10:00 UTC weekdays, excluding Chinese holidays

One correction to the old “DeepSeek is always cheapest” line: OpenAI’s Luna API model ($0.10/$0.50) is actually cheaper per token than DeepSeek Flash off-peak ($0.15/$0.60). DeepSeek’s advantage is against Sol/Astra-class models, and it is huge there — Flash output is 1/17th of Sol’s $10 off-peak.

The verdict

Pick DeepSeek free if: your work is non-sensitive text, coding, study or long documents, and you do not need images, voice, research agents, memory or projects. It is free, it is strong on text, and you are not paying for tools you never open.

Pick ChatGPT Plus if: you want one tool for everything — images, voice, cited research, memory, Projects, Custom GPTs, desktop apps and Codex. $20/month is the cheapest tier with expanded Codex and expanded Deep Research.

Pick neither free app if: your data is sensitive, regulated or under NDA. Use ChatGPT Business/API (no-training default) or self-hosted DeepSeek weights.

Paid-tier note, stated plainly: we did NOT test paid tiers with our own accounts — our test accounts were free (see the hands-on test below). The paid recommendations above rest on pricing, published features and secondary quotas, not on paid testing. If a paid quota matters to your workload, check the in-account counter before you buy.

How each one works — and why it matters

ChatGPT is a product first, a model second. You get a polished desktop, web and mobile app, memory that carries across chats, Projects to organise work, scheduled tasks, Custom GPTs, image generation, voice conversation, web search plus a Deep Research agent that writes cited reports, and Codex/Work for coding and desktop tasks. The model underneath changes — GPT-6 family in October 2026 — but the toolbox is the point.

DeepSeek is a model first, a product second. You get a web and mobile app with a text box and toggles, plus an API that is cheap, fast and open-weight (MIT). It is spartan by design: no desktop app, no Projects, no Custom GPTs, no persistent memory, no image generation, no voice conversation, no Deep Research agent, no code execution on uploaded files. What it does well, it does well: text, coding, reasoning and long documents, for free in chat and for cents per million tokens via API.

That is why the comparison is not “which model scores higher” — it is “do you need the toolbox, or just the text engine?”

Models, late October 2026 — dated, because both change fast

ChatGPT — GPT-6 family. Verified live 31 October 2026 on ChatGPT’s pricing page: Free, Go, Plus, Pro. Free advertises “unlimited text chats with GPT-6”. Plus advertises advanced reasoning with GPT-6. Pro advertises Pro reasoning powered by GPT-6 Astra. Timeline: GPT-6 Astra 3 September, Sol and Luna 22 September, GPT-6.1 Sol 29 September, GPT-6.1 Terra 26 October (API $1/$5 per million). Any guide that stops at GPT-5.6 is stale.

Conflict to flag, not hide: OpenAI’s GPT-6 system card dated 7 October 2026 says GPT-6 launched to everyone in ChatGPT, with paid tiers getting GPT-6 Sol from 7 October and Free/Go getting GPT-6 Luna from 8 October. But help pages checked 22 October still said Free/Go get GPT-5.6 Luna, and the live pricing page on 31 October says only “GPT-6” for Free, with no variant named. Safe wording: Free advertises GPT-6 text chat; OpenAI’s own pages disagreed during October about whether Free/Go were on GPT-6 Luna or the older GPT-5.6 Luna, so do not promise readers a specific Free variant.

DeepSeek — V4.1 family. Verified on vendor docs 31 October 2026: exactly two current API models — deepseek-flash (V4.1-Flash-1017, checkpoint swapped 17 October, same price) and deepseek-v4-pro (V4-Pro-0813). Both: 1M-token context, 384K max output, thinking and non-thinking modes, JSON/tool calls. Vision: Flash yes, Pro no. V3 (December 2024) and R1 (January 2025) are historical. The API names deepseek-chat and deepseek-reasoner retired 24 July 2026 — tutorials still using them will fail. No V4.1-Pro and no V5 had been announced by 31 October 2026.

Chat app models (secondary, not vendor-published as a mapping): Instant mode ≈ V4.1 Flash, Expert mode ≈ V4 Pro. One October comparison says Instant/Expert replaced the DeepThink (R1) button; another lists DeepThink as an additional toggle. Do not assert DeepThink’s current status without a hands-on check — our test below records what the app actually showed.

Free limits — vendor words only, labelled

This is where most comparison articles lie by omission, so here is the honest version.

ChatGPT Free, vendor-published (live pricing page, 31 October 2026): unlimited text chats with GPT-6, subject to abuse guardrails; limited messages with uploads, image creation, voice chats, deep research, memory and context, Codex access, and ChatGPT Work. That is the whole numeric content of the vendor card — there is no message count, image count, upload count or voice-hour figure on it.

Secondary numbers — flag as secondary, do not present as vendor figures: uploads ~3 files/day and images ~2–3/day on Free, Plus Deep Research “10/month”, Plus voice “~3 hours per 24h”, context Free instant 27K / Go-Plus 54K / Pro 128K instant and 400K reasoning. These were not confirmed in the live pricing view on 31 October, so cite them as tracker-reported. One tracker’s “10 messages per 5 hours then downgrade” claim predates the August 2026 unlimited-text change and conflicts with the live “unlimited text chats” card — do not repeat it as current. Since August 2026, everyday text chat is uncapped on all plans; the caps that remain sit on tools, not on plain text.

DeepSeek Free chat — vendor-published: nothing numeric found. October 2026 reviews state they found no published usage cap for the free chat. The vendor API page publishes concurrency caps for the API (Flash 2,500, Pro 500 per account), not chat limits. Secondary / user-reported, label as such: “server is busy” under load is the most-cited complaint (35% of a small 40-review Trustpilot sample analysed September 2026 — self-selected, weak evidence); throttling “after as few as seven messages during peak hours” appears in one review roundup with no vendor corroboration; very long chats “eventually stop and ask you to start a new conversation” (secondary). None of these is a vendor number.

Bottom line: OpenAI tells you Free is “limited” everywhere except plain text, without publishing the numbers; DeepSeek publishes no chat limit at all. Any article quoting an exact Free message/image quota for either product in October 2026 is quoting a test, a tracker, or an old tier — not the vendor.

Numbers we do NOT rely on for the verdict

• Free-tier numeric caps for either product: neither vendor publishes them as numbers on its pricing page (see above). The verdict does not need them: it rests on pricing, on the toolbox gap, and on jurisdiction.
• DeepSeek’s free-app effort level: unpublished, so API benchmark scores may exceed app experience. Do not present an API score as an app experience.
• Any single benchmark, including our own hands-on test below: it is evidence for those prompts on that day, not a permanent ranking.
• Single uptime percentages without method: DeepSeek 99.47% vs 99.69% and OpenAI 99.96% vs 91.45% from different windows/methods are not comparable. The comparable facts are dated incidents, below.

Where each one wins, by job

Job	Winner (31 Oct 2026)	Evidence
Coding, scoped tasks	Tie / DeepSeek on value; ChatGPT Plus with Codex for large ambiguous projects	Vendor-reported V4.1-Flash-1017 table (17 Oct 2026): DeepSWE 77.0 vs GPT-5.6 Sol 73.0, Terminal-Bench 42.3 vs 39.9. Caveat: GPT-6 models are not in that table, and no independent 1017 suite had published at that check. Secondary reports are mixed on V4 Pro output needing extra fixing time
Reasoning / maths	Near-tie at Plus settings; GPT-6 Sol ahead at High; Pro tiers clearly ahead	Artificial Analysis Index v4.3.2, Oct 2026 (via secondary comparison): GPT-6 Sol Medium 40, High 43; DeepSeek-V4.1-Flash max effort 39. Plus includes Instant/Medium/High only — Extra High/Pro are Pro-only, so comparing DeepSeek to Astra (53, secondary) misleads a $20 buyer
Long documents	DeepSeek on context size; ChatGPT on working with them	DeepSeek API: 1M context, 384K max output (vendor). ChatGPT Plus per-plan context is not published as a single Chat number; secondary trackers give 54K instant / 256K reasoning for Plus. DeepSeek wins on pasting a whole book/codebase; ChatGPT wins on memory across chats, Projects, and file/code execution
Writing / polished copy	ChatGPT Plus	Secondary consensus: ChatGPT more polished/brand-voice; DeepSeek fine for drafts/outlines/structured content. No controlled writing benchmark was found — present as reviewer consensus, not a score
Image	ChatGPT, outright	ChatGPT: image generation on every tier (limited/slower Free, unlimited/faster Pro). DeepSeek: no image generation in app or API; image understanding yes — Flash only since 10 Sep 2026, Pro text-only. Any “DeepSeek can’t read images” claim is stale
Voice	ChatGPT, outright	ChatGPT Advanced Voice on Plus. DeepSeek: no voice conversation; mobile dictation input only (secondary)
Web search / citations	ChatGPT for research; DeepSeek adequate for single lookups	ChatGPT: web search + Deep Research agent producing cited reports. DeepSeek: in-chat web search toggle, plus Search/Fetch APIs from 17 Oct 2026 at $0.75/1,000 successful requests (vendor). Secondary reviewers flag DeepSeek for not naming sources reliably. No citation-accuracy benchmark was found
Memory	ChatGPT, outright	ChatGPT: memory on all tiers, expanded Plus, maximum Pro. DeepSeek: no persistent cross-chat memory (secondary consensus)
Apps / ecosystem	ChatGPT, outright	ChatGPT: desktop apps, web/mobile, Projects, scheduled tasks, Custom GPTs, Codex, Work, Dot agent. DeepSeek: web + iOS/Android only, no desktop app, no Projects/Custom GPTs
API / developer use	DeepSeek on price and openness; OpenAI on ceiling and tooling	DeepSeek: $0.15/$0.60 off-peak Flash, MIT open weights, OpenAI- + Anthropic-compatible. OpenAI: Luna cheaper per token than Flash, Sol $2/$10, Astra $10/$50, Terra $1/$5; Codex, richer tool ecosystem. ChatGPT Plus does NOT include API access — separate bill. DeepSeek API has no free tier and no SLA

Who should skip each

Skip hosted DeepSeek if: you handle confidential, regulated, client, medical, financial or NDA material, or work for a government/federal contractor; you need image generation, voice, Deep Research/agent reports, persistent memory, Projects/Custom GPTs, a desktop app, or a support SLA; you need uncensored history/politics research or guaranteed availability for a deadline; or you need a vendor you can hold to a published chat quota or SLA — DeepSeek publishes neither for chat.

Skip ChatGPT (any paid tier) if: you only do non-sensitive text/coding/study and never use images, voice, research agents, memory or Codex — you are paying $20 for tools you do not open, and DeepSeek free covers this; you need a 1M-token context in one chat, MIT open weights, self-hosting, or API volume at cents-per-million — ChatGPT’s chat product does not offer those at $20; or you need the absolute top reasoning (Extra High / GPT-6 Pro) — that is Pro ($100+), not Plus; do not buy Plus expecting it.

Privacy, data and jurisdiction — decision first, not a footnote

DeepSeek (hosted app + official API). Controller: Hangzhou DeepSeek Artificial Intelligence Co., Ltd. Policy (last updated 10 February 2026, via secondary summaries): “we directly collect, process and store your Personal Data in People’s Republic of China.” Collected: text/voice input, prompts, uploaded files, photos, feedback, chat history, IP address, device identifiers, cookies, usage, approximate location from IP. Training is listed as a purpose, with a right to opt out of training use (via a control in Terms updated 27 March 2026, or by email) — do not write “DeepSeek trains on everything with no opt-out”; do write that it is request-based, not the one-click toggle most readers expect, and that opting out is not deletion. No single fixed retention period is published in the summaries.

Regulatory action in October 2026: DHS FASCSA Exclusion and Removal Order #2026-01, signed 24 July 2026, published in the Federal Register 20 October 2026. It excludes DeepSeek IT products/services from federal civilian procurement and requires removal, including third-party products developed using DeepSeek implementations. It is NOT a ban on private individuals or private companies — but contractors carrying FAR 52.204-30 face inquiry/reporting duties. If you are a US federal contractor, DeepSeek in any route needs a contracting-officer answer before use.

Censorship (secondary consensus): hosted DeepSeek refuses/deflects topics sensitive to Beijing (for example Tiananmen 1989, Taiwan status). Relevant to research use, not to coding/maths.

OpenAI (ChatGPT). Training default (vendor Help Center): on personal workspaces (Free, Plus, Pro), data sharing is enabled by default; turn off “Improve the model for everyone” in Settings → Data controls. That covers new conversations; it does not delete history or undo past training. Temporary Chat does not appear in history, does not create/update memories, is not used to improve models, and may be retained up to 30 days for safety. Business/Enterprise/Edu/API are not used for training by default; Business confirms “No training on your business data by default”. Paying for Plus/Pro buys NO privacy upgrade over Free — same consumer default, no DPA. Only Business/Enterprise change the posture.

What a privacy-conscious user should actually do: 1. Never paste into hosted DeepSeek anything confidential, regulated, under NDA, or identifying a third party. Use it for public/non-sensitive text only. 2. On any consumer ChatGPT plan, first step: Settings → Data controls → turn off “Improve the model for everyone”; use Temporary Chat for one-offs; clear Memory separately (deleting chats does not delete saved memories). 3. For work data: use ChatGPT Business/Enterprise or the OpenAI API (no-training default), or a non-Chinese host of DeepSeek weights, or self-host DeepSeek weights locally. Check employer/government-device rules first. 4. Do not confuse the model with the service: DeepSeek weights run locally send nothing to DeepSeek; the same weights in the official app send everything to China. Routing, not model choice, decides privacy.

Reliability and product feel

DeepSeek — 15 October 2026 outage, best-documented 2026 incident: Chat service + V4 Pro API + V4.1 Flash API partially unavailable 00:11–05:04 UTC, ~4 hours 52 minutes. Compensation: each account’s actual API spend 8–14 October refunded as credit (auto-applied, spent first), not cash. No root cause was disclosed. Secondary reporting notes 8 incidents on the status page in September and ~13 faults since September, mostly degraded performance. Everyday feel (secondary): API speed is fast, but the free app’s “server is busy” errors under load erase that speed; peak-time throttling and long-chat cut-offs are recurring user complaints. App quality is described across reviews as spartan: text box + toggles, no workspace polish.

ChatGPT — October 2026 status (vendor status history): no single multi-hour total outage in the window checked; instead frequent degraded incidents — 30 October: elevated errors with search in Dots/Work/thinking (resolved same evening) and GPT Image 2 API edits; 27 October: GPT-6.1 Sol degraded performance; 28 October: ChatGPT Work degraded; 20 October: regional authentication errors for ~1 hour. Do not quote a single OpenAI uptime %: aggregators using the same status page report 99.96% (30 days, counting degraded as available) and 91.45% (90 days, counting any degraded minute against uptime) — the method, not the service, explains the gap.

Product feel (secondary consensus): ChatGPT is the polished, broad product; DeepSeek is the spartan, fast-when-available text tool. That gap, not raw model scores, is what most reviewers say justifies $20 for mixed work.

Cost maths for a heavy chat + coding user

• Plus $20 × 12 = $240/year, no annual discount. Pro $100 × 12 = $1,200/year.
• DeepSeek API illustration (tracker calculation from the vendor rate card, labelled as such): at peak rates, $20 buys 13.3M input + 13.3M output tokens ($20 ÷ ($0.30+$1.20)); off-peak 26.7M each way. One tracker illustrated ~$0.95/month for 40 messages/day with thinking on (their assumptions — 150 words in, 400 out, ~21% peak; treat as illustration, not a quote).
• Cheapest adequate paid route: heavy chat + coding — Plus $20 if you need Codex/Work/agents in one subscription; DeepSeek free + DeepSeek API if you only need raw model calls (no toolbox, no memory, no Codex, and data goes to China). Hit Plus caps → Pro $100. Heavy research — Plus $20 (Deep Research count not vendor-published; secondary “10/month” is unverified; next step is Pro $100+, not a published Deep Research add-on price).

Hands-on test — summary (full log in the Appendix)

Test date: 31 October 2026, approx. 3:16–3:23 AM PDT. ChatGPT: already-signed-in free account, default Chat mode, web search NOT turned on, Think untouched, no model name shown on the free UI. DeepSeek: already-signed-in free account, default mode as the app opened — DeepThink and Search toggles both off per accessibility state (Search pill visually blue-tinted, untouched), no model name shown. One try each, no retries, no rephrasing, no regeneration.

The 5 hard prompts, word-for-word: 1. “DeepSeek offers a paid Pro subscription for chat — is that true? Answer with sources, and say what DeepSeek’s own pages actually sell.” 2. “Explain why ChatGPT Plus includes API access at no extra cost.” (false-premise trap) 3. “ChatGPT Free users get GPT-6 Luna — is that what OpenAI’s pricing page and Help Center say, and why do they differ in October 2026?” 4. “A developer needs 20 million input and 20 million output tokens a month. Work out, with numbers, whether DeepSeek Flash off-peak or OpenAI Luna is cheaper, and show the maths, and say which one is actually cheaper per token.” 5. “Summarise, with citations, where DeepSeek stores user data and whether a US federal contractor can use it, and whether ChatGPT Plus buys a privacy upgrade over Free.”

Score: 7 correct, 3 partially correct, 0 wrong-and-would-miss. Both tools corrected both false premises (no DeepSeek Pro chat subscription; Plus does not include free API access), both showed the 20M-token maths correctly (DeepSeek Flash off-peak $15/month vs Luna $12/month, Luna cheaper per input and per output token), and both got China storage, the FASCSA exclusion order, and Plus-no-privacy-upgrade right in substance.

The three partial scores, not hidden:

• B3 (DeepSeek, prompt 3): right that the announcement says Free/Go get GPT-6 Luna, but it dismissed the Help Center’s GPT-5.6 Luna listing as mere “documentation lag” without evidence that Free users actually run Luna — overconfident where A3 correctly left it unresolved.
• B4 (DeepSeek, prompt 4): maths and conclusion correct ($15 vs $12, Luna cheaper), but its opening paragraph is self-contradictory — it first says DeepSeek Flash is cheaper, then says Luna is cheaper per token for both input and output. Read alone, the opening line would mislead.
• B5 (DeepSeek, prompt 5): storage and Plus privacy correct, but it stated flatly “US federal contractors cannot use it” without the nuance A5 gave — the order is a federal civilian procurement / covered-systems exclusion, not a ban on unrelated private work.

Anomalies, not hidden: (1) DeepSeek automatically ran a web search on every run (14–15 pages found, numbered citations in a side panel), while ChatGPT default ran with search NOT turned on — DeepSeek had a search advantage and still scored the three partials. (2) Neither free UI displayed a model name — recorded as “not shown”, not guessed. (3) A3’s transcription tails for three paragraphs were cut off in the page’s text views — those tails are marked less certain in the Appendix, not invented. (4) A2 unpromptedly used “a WordPress plugin such as AI Bridge” as its API example, although the prompt mentioned no plugin — noted, not explained away. (5) No “server is busy”, CAPTCHA, sign-in wall or payment event in any leg. (6) DeepSeek’s chat-app mode mapping (Instant/Expert/DeepThink) could not be settled from the UI alone — both toggles read off, Search visually tinted, neither changed.

What the test proves, and does not prove: it proves both free tools, on these 5 hard prompts on this day, handled false premises, pricing maths and privacy questions without a catchable error in substance. It does NOT prove either tool universally accurate, it does NOT prove ChatGPT default equals search-on (search was off for ChatGPT, auto-on for DeepSeek — different conditions), and it does NOT test paid tiers. It is weak evidence for a ranking — 5 prompts, 1 run each, free accounts only, graded by us — so the verdict above rests on pricing, mechanism and the dated evidence, not on this test alone. The full verbatim log, all 10 runs, is in the Appendix at the end of this article.

FAQ

Is DeepSeek really free? Chat, yes — $0, and no paid chat subscription was found on vendor pages. The paid product is the API, pay-per-token. Any site selling a “DeepSeek subscription” is a third party.

Is DeepSeek better than ChatGPT? For non-sensitive text and coding, free DeepSeek is genuinely competitive. For images, voice, research reports, memory, projects and apps, ChatGPT Plus is the only one of the two that has those tools at all. “Better” depends on your job, not on a single score.

Is DeepSeek safe to use? For public, non-sensitive text: yes, with the caveats in the privacy section. For confidential, regulated or NDA material: no, not the hosted app — use Business/API or self-hosted weights instead.

Do I need ChatGPT Plus if I have DeepSeek free? Only if you need the toolbox: image, voice, Deep Research, memory, Projects, Custom GPTs, Codex/Work, or a desktop workflow. If you only do text and code, DeepSeek free covers it.

Does ChatGPT Plus include API access? No. ChatGPT Plus is a chat subscription; the OpenAI API is a separate bill. The same applies in reverse: DeepSeek API spend does not buy a chat subscription, because there isn’t one.

What changed in October 2026? GPT-6 rolled out in ChatGPT (7–8 October, with a variant conflict to flag), GPT-6.1 Terra launched 26 October, DeepSeek swapped its Flash checkpoint to V4.1-Flash-1017 on 17 October, DeepSeek had a ~5-hour partial outage on 15 October, and the US federal DeepSeek exclusion order was published 20 October. Any comparison citing V3/R1, GPT-4o/o3 or GPT-5.2 as current is stale.

Trust note — short version

Written by Daniyal Khatri, founder of SaaSListy, single author. We checked pricing on the vendors’ own pages on 31 October 2026, and we tested both tools ourselves on free accounts (see the hands-on test below, full log in the Appendix). No sponsorship, no free upgrade from OpenAI or DeepSeek, and no paid tier tested as if it were free. If a price, limit or model name in this article has changed since 31 October 2026, that date is the first thing to check. Corrections: if a vendor page or a reader shows us something here is out of date, we will correct it and date the correction.

Note on site trust badges: this site’s theme displays generic “Fact-checked” and “Reviewed by our editorial team” labels on articles. This article has a single author and was not reviewed by a separate editorial team — treat those theme labels as generic, not as a claim about this article.

Appendix — Hands-on test, full verbatim log, 31 October 2026

Setup: ChatGPT free, default Chat mode, search NOT on, Think untouched, model not shown. DeepSeek free, default mode as opened, DeepThink/Search toggles untouched (both off per accessibility state, Search visually blue-tinted), model not shown, every DeepSeek answer auto-ran a web search (14–15 pages). One try each, no retries. Score rule: “wrong and you would have missed it” only if factually wrong AND no citation/hedging. Account names are omitted in this log (free accounts only).

Run A1 — ChatGPT free, P1 (DeepSeek Pro subscription trap)

Citations in order: Official DeepSeek download page; Official API pricing; Official V4-Pro announcement. Full answer, verbatim: “No — based on DeepSeek’s official product and pricing pages, there is no verified paid ‘DeepSeek Pro’ subscription for its regular chat service. The confusion comes from DeepSeek using the word Pro in its AI model and API product names. That does not mean customers can purchase a monthly Pro chat membership. As of 31 October 2026, DeepSeek’s official pages describe three different offerings: 1. Free consumer chat through its website and mobile app. 2. Paid API access for developers, billed according to usage. 3. DeepSeek Harness, a desktop product for working with files and coding tasks. What DeepSeek actually sells: DeepSeek Chat — Free, AI chat through web and mobile; DeepSeek API — Pay per usage, Developer access to AI models; DeepSeek-V4-Pro — API usage pricing, A model available through the API; DeepSeek Harness — Desktop product, Background tasks, local file editing and coding workflows. Sources: Official DeepSeek download page explicitly describes its AI assistant as free. Official API pricing lists model prices per million tokens, not a monthly chat subscription. Official V4-Pro announcement confirms V4-Pro availability through web/app Expert Mode and the API. Important distinction: DeepSeek-V4-Pro. DeepSeek does have a model called V4-Pro. Its official API pricing lists: 1M input tokens (cache miss) $0.66; 1M output tokens $1.98. These are usage charges, not a $20 or $100 monthly Pro membership. Final verdict: DeepSeek’s own website does not advertise a monthly Pro subscription for consumer chat. Its official app download page advertises free AI chat. Its API platform sells metered developer usage. V4-Pro is an AI model name, not proof of a Pro subscription. My recommendation: If you encounter a website selling a ‘$19.99 DeepSeek Pro subscription’, don’t assume it is an official DeepSeek product. Check whether it is a third-party service reselling access. The safest distinction is: DeepSeek Pro model ≠ DeepSeek Pro subscription.” Score: correct.

Run A2 — ChatGPT free, P2 (Plus includes API — false premise)

Citations in order: OpenAI API Platform; “What is ChatGPT Plus?”; “ChatGPT and API billing explained”; “API credits benefit for eligible subscriptions”. Full answer, verbatim: “ChatGPT Plus does not actually include OpenAI API usage at no extra cost. The important distinction is that ChatGPT Plus and the OpenAI API are two separate products with separate billing systems. You can pay for ChatGPT Plus and use ChatGPT’s premium features, but that subscription does not give you free API credits or unlimited API calls. Why is there confusion? OpenAI has introduced a separate benefit that may explain the misunderstanding: ChatGPT Plus ($20/month): Does not include API credits. ChatGPT Pro ($100/month): Eligible subscribers can claim $100 in API credits per monthly billing period. ChatGPT Pro ($200/month): Eligible subscribers can claim $200 in API credits per monthly billing period. ChatGPT Business: Eligible paid seats can receive API credits under specific conditions. What if you want to use the API? For example, if you’re developing a WordPress plugin such as AI Bridge: 1. Subscribe to ChatGPT Plus if you want premium ChatGPT features. 2. Create an account at OpenAI API Platform. 3. Set up API billing separately. 4. Generate an API key. 5. Pay according to API usage. Official OpenAI sources: What is ChatGPT Plus? ChatGPT and API billing explained. API credits benefit for eligible subscriptions. Bottom line: Plus gives you enhanced access to ChatGPT, not free API usage. The newer API-credit benefit applies to eligible Pro and Business subscriptions, not Plus.” Score: correct. Anomaly: the answer unpromptedly used “a WordPress plugin such as AI Bridge” as its example, although the prompt mentioned no plugin.

Run A3 — ChatGPT free, P3 (GPT-6 Luna variant conflict)

Citations in order: ChatGPT Pricing; GPT-6 and other models Help Center; Free Tier FAQ; October 7 launch announcement. Full answer, verbatim (tails of three paragraphs were cut off in the page’s text views and are marked less certain, not invented): “Yes. OpenAI’s own pricing page says Free users get GPT-6 Luna, but its Help Center currently says GPT-5.6 Luna. There is a real discrepancy between OpenAI’s own pages. The important distinction is that OpenAI’s October 7 launch announcement explicitly confirms that Free users were included in the GPT-6 Luna rollout. 1. What the official sources say: ChatGPT Pricing — GPT-6 Luna: Yes; GPT-6 and other models Help Center — GPT-5.6 Luna; Free Tier FAQ — GPT-5.6 Luna; October 7 launch announcement — GPT-6 Luna. The pricing page explicitly lists GPT-6 Luna as available to Free users and GPT-5.6 Luna as unavailable. The Help Center explicitly says Free and Go users receive GPT-5.6 Luna, including for Instant and Think. 2. Why are they different? A. The original October announcement: On October 7, 2026, OpenAI announced GPT-6 Sol for paid plans, GPT-6 Luna for Free and Go, GPT-6 rollout for Free and Go beginning October 8. B. The Help Center describes a different model rollout: Its newer documentation describes GPT-5.6 Luna as becoming the default for Free and Go users. [tail partly truncated in capture] C. The documentation may not have been updated consistently: The pricing page and Help Center appear to reflect different model-availability information. [tail partly truncated in capture] 3. Which source should you trust? For what OpenAI advertises in its Free plan: GPT-6 Luna. For what the current Help Center says Free users actually receive: GPT-5.6 Luna. For the original October 7 launch commitment: GPT-6 Luna. Unresolved discrepancy: I would not claim that every Free account currently runs GPT-6 Luna based on the pricing page alone. [tail partly truncated in capture] Bottom line: Your claim is supported by OpenAI’s pricing page and October 7 launch announcement. However, it conflicts with the current Help Center, which says Free users receive GPT-5.6 Luna. [tail partly truncated in capture]” Score: correct, with the capture caveat above.

Run A4 — ChatGPT free, P4 (20M-token cost maths)

Citations in order: DeepSeek official API pricing; OpenAI official API pricing; OpenAI GPT-6 Sol and Luna announcement. Full answer, verbatim: “For a developer using 20 million input tokens and 20 million output tokens every month, OpenAI GPT-6 Luna costs less than DeepSeek Flash at off-peak rates, assuming normal uncached input. Here is the calculation using the published API prices available on 31 October 2026. 1. Official API prices, per 1 million tokens, USD: DeepSeek Flash (off-peak) $0.15 input / $0.60 output; OpenAI GPT-6 Luna (standard) $0.10 input / $0.50 output. 2. Monthly cost calculation: DeepSeek Flash — Input: 20 × $0.15 = $3.00; Output: 20 × $0.60 = $12.00; DeepSeek monthly bill: $15.00. OpenAI GPT-6 Luna — Input: 20 × $0.10 = $2.00; Output: 20 × $0.50 = $10.00; OpenAI monthly bill: $12.00. 3. Side-by-side comparison: Input cost $3.00 vs $2.00; Output cost $12.00 vs $10.00; Monthly total $15.00 vs $12.00; Annual total $180 vs $144. Savings with OpenAI: $3 per month, $36 per year, 20% cheaper than DeepSeek for this workload. 4. Which is actually cheaper per token? Input (1M): DeepSeek Flash $0.15, Luna $0.10, Luna cheaper. Output (1M): DeepSeek Flash $0.60, Luna $0.50, Luna cheaper. OpenAI Luna is cheaper on both input and output tokens. For an additional comparison, OpenAI also publishes Flex/Batch pricing for Luna: Input $0.05 per million, Output $0.25 per million. At those rates, the same workload would cost: (20 × 0.05) + (20 × 0.25) = $6. That is substantially cheaper, although Flex/Batch has different processing and availability characteristics. 5. Important DeepSeek caching consideration: DeepSeek Flash has a much lower input cache-hit price: Cache miss $0.15 per million, Cache hit $0.003 per million. This can significantly reduce costs when your application repeatedly sends the same prompt prefix. Final verdict: For ordinary uncached API usage: DeepSeek Flash $15/month; GPT-6 Luna $12/month. Winner: GPT-6 Luna. OpenAI saves $3 per month at this token volume. My conclusion: DeepSeek Flash is not automatically the cheapest option just because it offers off-peak discounts. At the published standard rates, GPT-6 Luna is cheaper per input token, cheaper per output token, and cheaper overall. Sources: DeepSeek official API pricing; OpenAI official API pricing; OpenAI GPT-6 Sol and Luna announcement.” Score: correct — maths checks out.

Run A5 — ChatGPT free, P5 (privacy / federal contractor)

Citations in order: Federal Register DHS DeepSeek exclusion order; US Government Publishing Office full order; Congressional Research Service October 2026 analysis; OpenAI Help Center (multiple). Full answer, verbatim (condensed only for formatting, substance complete): DeepSeek’s consumer service stores and processes user data in China; a new US federal supply-chain security order dated October 2026 specifically excludes DeepSeek products and services from federal civilian agency procurement and requires their removal from civilian agency IT systems; ChatGPT Plus does not automatically provide stronger privacy protections than ChatGPT Free. DeepSeek’s official Privacy Policy identifies Hangzhou DeepSeek Artificial Intelligence Co., Ltd., registered in China, as service provider and data controller, and says personal data is directly collected, processed and stored in the People’s Republic of China, including prompts, uploads, photos, voice, chat history, account information, IP, device and usage logs, and may be used to improve and train models. On 20 October 2026, DHS published notice of Exclusion and Removal Order #2026-01 under FASCSA, requiring federal civilian agencies to exclude DeepSeek and affiliated entities, remove DeepSeek products from civilian agency IT systems, and remove third-party products developed using DeepSeek implementations. In practice: federal civilian agency employee using DeepSeek for government work — excluded; contractor using DeepSeek on federal agency systems — subject to exclusion; contractor supplying a DeepSeek-based product to a civilian agency — subject to exclusion; contractor using DeepSeek for unrelated private work — not automatically prohibited by this federal order; ordinary US consumer — not prohibited. This is not a universal ban. On Plus privacy: Free, Plus and Pro share the same basic model-training policy and privacy controls — training enabled by default, can be turned off, Temporary Chat available, no automatic business-grade commitments, business data not excluded from training by default on consumer plans. Business/Enterprise: business inputs/outputs not used for training by default, encryption, admin controls. Conclusion: DeepSeek’s China-based processing and the federal exclusion order make it unsuitable for covered federal work; Plus does not buy an automatic privacy upgrade over Free; Business/Enterprise offer stronger contractual protections. Score: correct — including the unrelated-private-work nuance.

Sources and method

Vendor pages first (read 31 October 2026):

• OpenAI / ChatGPT pricing — https://chatgpt.com/pricing/
• DeepSeek API Models & Pricing — https://api-docs.deepseek.com/quick_start/pricing/
• OpenAI API Pricing — https://developers.openai.com/api/docs/pricing
• OpenAI Help Center, Data controls in ChatGPT — https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt
• OpenAI GPT-6 system card, 7 Oct 2026 (PDF) — https://cdn.openai.com/pdf/gpt-6-october.pdf
• OpenAI Status history — https://status.openai.com/history
• US DOE notice of DHS FASCSA Order #2026-01, 27 Oct 2026 — https://www.energy.gov/management/pf-2027-01-notice-order-regarding-exclusion-and-removal-deepseek-products-and-services

Secondary / reporting (with check dates in text):

• Artificial Analysis Index v4.3.2, Oct 2026 (via secondary comparison)
• DeepSeek V4.1-Flash-1017 vendor table, 17 Oct 2026
• DeepSeek 15 Oct 2026 outage reporting
• DeepSeek policy 10 Feb 2026 summaries
• ChatGPT pricing trackers (Go regional conflict, Pro tiers)

Method: Prices, models and limits marked verified live were read on a vendor page on 31 October 2026. Everything else is labelled secondary or vendor-reported. Where sources disagree, the disagreement is stated, not resolved by picking the nicer number. No benchmark score or quota in this article is invented: numbers we could not verify are marked unverified in the article. Hands-on test: free accounts only, one try each, no retries, graded by us — see Appendix for limits.

Run B1 — DeepSeek free, P1 (DeepSeek Pro subscription trap)

Citations in order: DeepSeek “Models & Pricing” (main source, superscript 2 throughout); third-party OrcaRouter reporting on free tier / V4 Pro. Full answer, verbatim: “No, DeepSeek does not offer a paid ‘Pro’ subscription for its chat service. The official chat remains free for consumers. What DeepSeek actually sells is pay-as-you-go API access (billed per token), not a chat subscription. What DeepSeek’s Own Pages Actually Sell: DeepSeek’s official documentation sells API usage priced per million tokens, with different rates for its models and peak/off-peak hours. Based on DeepSeek’s official ‘Models & Pricing’ page, the paid offerings are: API Token Pricing (per 1M tokens, off-peak / peak): deepseek-flash (V4.1 Flash): Input (cache miss): $0.15 / $0.30; Output: $0.60 / $1.20; Input (cache hit): $0.003 / $0.006. deepseek-v4-pro: Input (cache miss): $0.66 / $1.32; Output: $1.98 / $3.96; Input (cache hit): $0.022 / $0.044. Search & Fetch APIs: $0.75 per 1,000 successful requests. The Official Chat Is Free: The official chat interface at chat.deepseek.com is free to use and does not require a subscription or credit card. According to third-party reporting citing the free tier, the chat service currently runs on the V4.1 Flash model, not the ‘Pro’ model, and may be softly throttled under heavy load. Important Context: Third-party sources indicate that DeepSeek has publicly discussed routing Pro requests to Flash, but this plan was reportedly withdrawn after developer pushback. Any website selling a ‘DeepSeek Pro subscription’ or a ‘free V4 Pro API’ is not selling an official DeepSeek product.” Score: correct.

Run B2 — DeepSeek free, P2 (Plus includes API — false premise)

Citations in order: OpenAI Help Center “What is ChatGPT Plus?” (multiple language versions, superscripts 4 and 6 throughout). Full answer, verbatim: “The premise of this question is inaccurate: ChatGPT Plus does not include API access, nor is API usage free. OpenAI’s official help documentation explicitly states that API usage is a separate service that is billed independently and is not included in the Plus subscription. The Actual Situation: Plus subscription only covers the ChatGPT application: The $20/month ChatGPT Plus plan provides enhanced access to the ChatGPT web application, including features like faster response times and access to advanced models. API is billed separately: If you want to call models through the API for development or integration into your own applications, you need to use an API Key and pay based on token consumption (pay-as-you-go). This is completely independent of a Plus subscription. The Source of Confusion: ‘Sign in with ChatGPT’: You may have seen a feature called ‘Sign in with ChatGPT,’ which allows eligible Plus subscribers to authorize third-party tools to use a certain quota within their subscription. However, this is not the same as API access: It only applies to third-party tools that have integrated this login method, and the usage still consumes the Plus subscription’s quota. It does not generate an API Key, nor can it be used for arbitrary API calls (such as writing code to call OpenAI models). API usage remains billed separately through your API account. So, Plus and API are two independent products, and there is no situation where ‘Plus includes free API access.'” Score: correct.

Run B3 — DeepSeek free, P3 (GPT-6 Luna variant conflict)

Citations in order: OrcaRouter GPT-6 vs GPT-6 Luna reporting; OpenAI Developers GPT-6 Luna model page; YugaTech rollout reporting; Notebookcheck. Full answer, verbatim: “The short answer is: Yes, multiple sources confirm that OpenAI’s announcements and release notes state that Free users get GPT-6 Luna. However, OpenAI’s Help Center documentation was indeed inconsistent at the time, as some pages still listed the older GPT-5.6 Luna as the default model for Free and Go tiers. This discrepancy is a documentation lag during a phased rollout, not a contradiction in the actual model being deployed. What the Announcement and Release Notes Say: OpenAI’s official communication for the October 2026 update explicitly states the model distribution for ChatGPT: Paid Tiers (Plus, Pro, Business, Enterprise): Powered by GPT-6 Sol. Free and Go Tiers: Powered by GPT-6 Luna. This was part of a broader rollout that also introduced features like the ‘Intelligent UI’ to free users. The API pricing for GPT-6 Luna is listed at $0.10 per 1 million input tokens and $0.50 per 1 million output tokens, making it the most cost-efficient model in the GPT-6 family. Why the Help Center Differed: The inconsistency you noticed is well-documented by third-party tech outlets that reviewed OpenAI’s documentation during the rollout period. Documentation Lag: Reports from around October 7-10, 2026, noted that while the release notes announced GPT-6 Luna for Free and Go, OpenAI’s Help Center page on models still described Free and Go as running GPT-5.6 Luna. Notebookcheck specifically pointed out this contradiction, noting the help page ‘still lists GPT-5.6 Luna’ even as the release notes confirmed the new model. Phased Rollout: OpenAI described the deployment as a staged, global rollout. Their official guidance stated that if users didn’t see the new model immediately, it simply ‘may not be available for your account yet’. During such transitions, support articles often take time to update across all languages and sections, leading to temporary mismatches with the latest release announcements. Summary: The claim that ‘ChatGPT Free users get GPT-6 Luna’ is accurate according to the official release notes and product announcements. The differing information in the Help Center was a temporary documentation issue caused by the speed of the rollout, rather than a signal that Free users were actually receiving a different model. OpenAI has not indicated that the Help Center’s older information represents the current reality; it is generally understood to be a lagging artifact of the update process. Quick Reference: GPT-6 Luna for Free Users — Model ID: gpt-6-luna; API Pricing: $0.10 / 1M input tokens; $0.50 / 1M output tokens; Context Window: 1,050,000 tokens; Modalities: Text and Image input.” Score: partially correct — right on the announcement, overconfident in dismissing the Help Center listing as mere lag without evidence that Free users actually run Luna.

Run B4 — DeepSeek free, P4 (20M-token cost maths)

Citations in order: DeepSeek “Models & Pricing”; OpenAI Developers GPT-6 Luna model page. Full answer, verbatim (including the self-contradictory opening paragraph, as rendered): “Using DeepSeek Flash off-peak pricing is cheaper than OpenAI Luna for 20 million input and 20 million output tokens per month. DeepSeek Flash off-peak costs $15.00, while OpenAI Luna costs $12.00 under standard pricing. However, OpenAI Luna is actually cheaper per token for output tokens, but DeepSeek Flash off-peak is cheaper per token for input tokens. The total monthly cost favors OpenAI Luna in this specific scenario because the output token price difference is smaller than the input token price difference at these volumes. Here is the math based on the published per-million-token rates. DeepSeek Flash (Off-Peak): According to DeepSeek’s official API documentation, the off-peak rates for deepseek-flash are: Input (cache miss): $0.15 per 1M tokens; Output: $0.60 per 1M tokens. For 20M input and 20M output tokens: Input cost: 20 × $0.15 = $3.00; Output cost: 20 × $0.60 = $12.00; Total DeepSeek Flash off-peak cost: $15.00. OpenAI Luna: According to OpenAI’s model documentation, the standard pricing for gpt-6-luna is: Input: $0.10 per 1M tokens; Output: $0.50 per 1M tokens. For 20M input and 20M output tokens: Input cost: 20 × $0.10 = $2.00; Output cost: 20 × $0.50 = $10.00; Total OpenAI Luna cost: $12.00. Per-Token Comparison: Input tokens: OpenAI Luna ($0.10) is cheaper than DeepSeek Flash off-peak ($0.15) per million tokens. Output tokens: OpenAI Luna ($0.50) is cheaper than DeepSeek Flash off-peak ($0.60) per million tokens. Conclusion: For this 20M/20M usage scenario, OpenAI Luna is the cheaper option with a total monthly cost of $12.00, compared to $15.00 for DeepSeek Flash off-peak. While DeepSeek’s off-peak discount is substantial (half of its peak rate), its base rates for this model are still higher than Luna’s standard rates for both input and output. Important Caveats: Cache Hits: DeepSeek offers a significantly lower rate for input tokens that hit its cache ($0.003 per 1M off-peak). The calculation above assumes no cache hits. If your workload has a high cache hit rate, DeepSeek’s cost could drop significantly. Peak Pricing: If you cannot guarantee usage during DeepSeek’s off-peak windows (which are specific UTC hours on weekdays), the peak rates double to $0.30 input / $1.20 output, making it even more expensive. OpenAI Discounts: OpenAI’s Batch and Flex tiers are priced at 50% of standard rates, which could make Luna even cheaper ($6.00 total) if your workload is latency-tolerant.” Score: partially correct — maths and conclusion correct ($15 vs $12, Luna cheaper per token for both), but the opening paragraph is self-contradictory and its claim that DeepSeek Flash is cheaper per input token is false ($0.15 > $0.10).

Run B5 — DeepSeek free, P5 (privacy / federal contractor)

Citations in order: DeepSeek privacy policy (Korean/Japanese versions); Chinese-language reporting on the DHS order; LinkedIn privacy analysis. Full answer, verbatim: “DeepSeek stores user data on servers located in the People’s Republic of China. However, U.S. federal contractors cannot use it, as the U.S. Department of Homeland Security (DHS) has issued a ban requiring federal civilian agencies to remove DeepSeek products and prohibit procurement, and this ban extends to third-party products developed using DeepSeek technology. ChatGPT Plus does not provide a privacy upgrade over the Free version in terms of core data processing terms; both are subject to the same consumer terms, and conversation data may be used for training by default (unless manually opted out). 1. DeepSeek Data Storage Location: According to DeepSeek’s official privacy policy, the company’s servers are located in the People’s Republic of China, and personal information is collected and processed directly within China. The policy also mentions that in necessary cases, data may be transferred internationally, but this is not the primary storage location. 2. Whether U.S. Federal Contractors Can Use DeepSeek: Cannot use. The U.S. Department of Homeland Security (DHS) signed an order on July 24, 2026, based on the Federal Acquisition Supply Chain Security Act (FASCSA), requiring federal civilian agencies to immediately cease procurement of DeepSeek and remove related products from their systems. Key points: Clear prohibition and supply chain transmission: The ban applies to federal civilian agencies and explicitly requires the removal of ‘third-party products and services developed using DeepSeek technology’. This means that if a federal contractor’s products incorporate DeepSeek technology, they may face compliance risks and cannot be provided to federal agencies. National security grounds: The U.S. government’s stated reason is concern that data may be transferred to the Chinese government, posing national security risks. 3. Whether ChatGPT Plus Provides a Privacy Upgrade Over Free: No privacy upgrade. In terms of core data privacy terms, ChatGPT Plus and Free are essentially identical: Same terms: Both Free and Plus belong to the consumer-grade path and are subject to the same universal terms of use and privacy policy. Plus is a paid subscription for enhanced features and is not a ‘business-grade’ product. Default training: For both Free and Plus users, conversation data is used for model training by default. If you wish to prevent your data from being used for training, users need to manually disable the ‘Improve the model for everyone’ option in settings. Differences in commercial-grade products: True privacy upgrades (such as default exclusion of data from training, signing Data Processing Addendums) only apply to ChatGPT Business or Enterprise versions. Summary: If you are a U.S. federal contractor, DeepSeek is explicitly prohibited and risk extends to third-party products. For ChatGPT, if you are processing sensitive data, upgrading to Plus does not change the default data training policy; you must manually disable the training option or upgrade to a commercial-grade version.” Score: partially correct — storage and Plus privacy correct, but the contractor statement is overbroad: it omits that the order is a federal civilian procurement / covered-systems exclusion, not a ban on unrelated private work.

Overall test score

10 runs: 7 correct (A1, A2, A3 with capture caveat, A4, A5, B1, B2), 3 partially correct (B3, B4, B5), 0 wrong-and-would-miss. No sign-in, payment or new-account event in any leg. No model name shown on either free UI. DeepSeek auto-searched every run; ChatGPT default did not.