ChatGPT vs DeepSeek

DeepSeek publishes its current models as open weights under the MIT license and sells a cheap API, but it stores data in China; ChatGPT's top models score higher on independent tests as of Nov 6, 2026 and offer images, voice and a training opt-out setting. For the cheapest capable API model the two are close, and OpenAI's GPT-6 Luna even undercuts DeepSeek's peak-hour price.

ChatGPT's best models score well above DeepSeek's on independent tests as of Nov 6, 2026: GPT-6 Astra reaches 57 on the Artificial Analysis Intelligence Index against 33 for DeepSeek V4.1 Flash. DeepSeek wins on openness: its current model's weights are free to download under the MIT license, while OpenAI's open models, gpt-oss, date from 2025. On privacy the two differ sharply: DeepSeek says it stores personal data in the People's Republic of China and offers training opt-out only by email, while ChatGPT has a setting. On price, the comparison is closer than its reputation: OpenAI's smallest current model, GPT-6 Luna, costs $0.10 per million input tokens, below DeepSeek Flash's $0.15 off-peak.

Learn more: DeepSeek on Ai.vi.ki.

Quick facts

Current models	ChatGPT: GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Luna; DeepSeek: V4.1 Flash, V4 Pro
Open weights	DeepSeek V4.1 Flash, MIT license, 763B parameters; OpenAI gpt-oss-120b and gpt-oss-20b (Aug 2025), Apache 2.0
Context window (API)	Both about 1 million tokens on current models
Cheapest API model	GPT-6 Luna $0.10 in, $0.50 out; DeepSeek Flash $0.15 in, $0.60 out off-peak ($0.30 and $1.20 at peak)
Consumer app price	ChatGPT Free, Go $8, Plus $20, Pro from $100; DeepSeek's site states no app price
Where data is kept	DeepSeek: China; OpenAI: regional data residency available to API customers
Availability	ChatGPT is not offered in China; DeepSeek's weights run anywhere

Prices per million tokens. Sources, as of Nov 6, 2026: OpenAI API pricing (including regional processing), OpenAI models, gpt-oss-120b, DeepSeek API pricing, DeepSeek V4.1 Flash on Hugging Face, DeepSeek privacy policy, ChatGPT supported countries, Codex pricing.

Verdict by task

Task	ChatGPT	DeepSeek	Verdict
Reasoning	GPT-6 Astra 57 on Artificial Analysis	V4.1 Flash 33	ChatGPT
Writing and chat	Best OpenAI model 22nd on Arena text (1485)	V4.1 Flash 35th (1476)	ChatGPT, narrowly
Coding	GPT-6 Astra 2nd on Arena WebDev (1773)	V4.1 Flash 16th (1665)	ChatGPT
Images	ChatGPT Images 2.6 on every plan	Reads images; no image generation documented	ChatGPT
Voice	Live voice with published hours	Not documented on the pages read	ChatGPT on the evidence
API price, small models	GPT-6 Luna $0.10 / $0.50	Flash $0.15 / $0.60 off-peak	ChatGPT's Luna is cheaper per token
Open weights	gpt-oss, from 2025	V4.1 Flash, MIT	DeepSeek
Privacy	Opt-out setting; US company	Data stored in China; opt-out by email	ChatGPT

Sources for the table: Artificial Analysis, Arena text leaderboard, updated Oct 27, 2026, Arena WebDev leaderboard, updated Oct 27, 2026, Images in ChatGPT, ChatGPT Voice, OpenAI API pricing, DeepSeek API pricing, DeepSeek privacy policy, all read Nov 6, 2026.

Reasoning, writing and coding

ChatGPT wins on every independent measure found, though by very different margins. The Artificial Analysis Intelligence Index, a set of reasoning, coding and agent tests, scores GPT-6 Astra 57, GPT-6.1 Terra 51 and DeepSeek V4.1 Flash 33. On Arena's blind-vote text leaderboard the gap almost vanishes: DeepSeek V4.1 Flash is 35th at 1476, level with GPT-6 Astra (37th, 1476) and a little behind OpenAI's best, GPT-5.6 Sol (22nd, 1485), as of Oct 27, 2026. For coding, Arena's WebDev board puts GPT-6 Astra 2nd at 1773 and DeepSeek-V4.1-Flash 16th at 1665, just behind GPT-6.1 Terra at 1673.

The fairest match is model size against model size. OpenAI's smallest, GPT-6 Luna, scores 28 on the Artificial Analysis index, below DeepSeek V4.1 Flash's 33. So for the price, DeepSeek's Flash does more than OpenAI's cheapest model, while OpenAI's mid-priced GPT-6.1 Terra does more than either. See OpenAI Benchmarks.

API price

Per token, OpenAI's cheapest current model is cheaper than DeepSeek's. GPT-6 Luna costs $0.10 per million input tokens and $0.50 output. DeepSeek prices by time of day: its Flash model costs $0.15 input and $0.60 output off-peak, and double at peak, which DeepSeek sets at 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays. DeepSeek's stronger V4 Pro costs $0.66 input and $1.98 output off-peak; OpenAI's GPT-6.1 Terra costs $1 and $5, and GPT-6 Astra $10 and $50.

Per finished task the picture shifts, because models use different numbers of tokens; Artificial Analysis measured GPT-6 Luna at $0.15 per index task but published no cost figure for DeepSeek in the rows read. Both offer about a million tokens of context on their current models. For OpenAI's full price list see OpenAI API Pricing.

Open weights

DeepSeek wins. Its newest model, DeepSeek-V4.1-Flash-1017, is on Hugging Face with 763 billion parameters and weights under the MIT license, a permissive license, so anyone can run it on their own hardware. Its V4 Pro weights were updated in Aug 2026. OpenAI's open models are gpt-oss-120b and gpt-oss-20b, released on Aug 5, 2025 under Apache 2.0, small enough that the 20b model runs in 16 GB of memory, and no 2026 successor appeared on OpenAI's Hugging Face page. Running a 763-billion-parameter model yourself needs data-center hardware, so most people meet DeepSeek through its app or API.

Images and voice

ChatGPT wins both on the evidence. ChatGPT Images 2.6 is on every ChatGPT plan, and Live voice gives 3 hours a day on Plus. DeepSeek's Flash model reads images as input (vision), but none of DeepSeek's pages read on Nov 6, 2026 describes image generation or a voice mode. See ChatGPT Images and Voice Mode.

Privacy

ChatGPT is the safer choice for most readers outside China. DeepSeek's privacy policy (last updated Feb 10, 2026) says it collects, processes and stores personal data in the People's Republic of China and uses it to train its models; it lists a right to opt out of training, exercised by emailing its privacy team rather than by a setting. OpenAI also trains on personal ChatGPT chats by default, but you turn it off under Settings > Data controls > Improve the model for everyone and keep your history; OpenAI's business plans are not used for training by default. Running DeepSeek's open weights on your own machine sends nothing to DeepSeek at all. More: Privacy and Data Controls and Is ChatGPT Safe.

Price of the apps

ChatGPT's consumer prices are published: Free, Go $8 a month, Plus $20 and Pro from $100. DeepSeek's home page names its current model but says nothing about the price of its chat app, and no paid consumer plan was found on its pages. ChatGPT Free and Go may show ads. ChatGPT is not available in China, Hong Kong and some other countries; the supported countries list has about 187 countries and territories. See ChatGPT Plans and Pricing.

When to choose DeepSeek, and when ChatGPT

Choose DeepSeek if you want open weights to run or fine-tune yourself under a permissive license, or a capable API model for bulk work in off-peak hours. Choose ChatGPT for stronger reasoning and coding, images and voice, published plans and limits, and a training opt-out you can flip in settings, and if you prefer your data not be stored in China. For bulk API work, test GPT-6 Luna as well: it is cheaper per token than DeepSeek Flash. More rivals: When to Use Something Else.

Learn more: DeepSeek on Ai.vi.ki.

Frequently asked questions

Is DeepSeek better than ChatGPT? Not on independent tests: GPT-6 Astra scores 57 on the Artificial Analysis Intelligence Index against DeepSeek V4.1 Flash's 33 (read Nov 6, 2026). DeepSeek is better if you need open weights, since its model is downloadable under the MIT license.

Is DeepSeek cheaper than ChatGPT? For the API, not always: OpenAI's GPT-6 Luna costs $0.10 per million input tokens and $0.50 output, against DeepSeek Flash's $0.15 and $0.60 off-peak and double that at peak. DeepSeek's V4 Pro, at $0.66 and $1.98 off-peak, is cheaper per token than OpenAI's GPT-6.1 Terra at $1 and $5.

Is DeepSeek safe to use? DeepSeek's privacy policy says it stores personal data in China and uses it to train its models, with opt-out by email. Running its open weights on your own hardware keeps your data local; ChatGPT lets you turn off training in Data controls.

Is DeepSeek open source and is ChatGPT? DeepSeek V4.1 Flash's weights are published under the MIT license. ChatGPT's models are not; OpenAI's open-weight models are gpt-oss-120b and gpt-oss-20b, released in Aug 2025 under Apache 2.0.

Can DeepSeek make images like ChatGPT? DeepSeek's Flash model can read images, but DeepSeek's pages describe no image generation as of Nov 6, 2026. ChatGPT creates images on every plan, including Free.