Belum ada versi bahasa Melayu, jadi dipaparkan dalam bahasa Inggeris.

DeepSeek vs OpenAI (November 2026): models, API pricing, benchmarks, open weights, and where your data goes

Quick answer

DeepSeek's API costs $0.30 in and $1.20 out per million tokens at peak for DeepSeek-V4.1-Flash-1017, and half that off-peak; OpenAI's GPT-6.2 Sol, released November 17, 2026, costs $2 / $10 and GPT-6.2 Astra $8 / $40 (each vendor's pricing page, read on November 18, 2026). DeepSeek publishes its weights under the MIT license and its privacy policy says it stores data in China; OpenAI's current models are closed, and its API offers data residency in the US, Europe, Japan, and 7 more regions. On the Terminal-Bench 4.1 leaderboard, GPT-6.1 Sol resolves 60.9% of tasks and DeepSeek V4.1 Flash 38.5%.

Publish my HTML page free

Free plan: no credit card, 1 published page, 50 form entries a month.

On this page

You're choosing an API for an app or a coding agent, and the two names on the shortlist are DeepSeek and OpenAI. They differ in three ways that decide most projects: the price per token, whether you can download the model, and which country your prompts are stored in.

Every fact below comes from DeepSeek's or OpenAI's own pricing page, model page, changelog, terms, or privacy policy, read on November 18, 2026 (Japan time). Benchmark numbers carry the name of whoever ran them. OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, so the OpenAI column is one day old.

DeepSeek vs OpenAI: the current models side by side

DeepSeek vs OpenAI, read on November 18, 2026. Prices are US dollars per million tokens, input / output, at each vendor's standard rate.

DeepSeekOpenAI
Models on the APIdeepseek-flash (DeepSeek-V4.1-Flash-1017) and deepseek-v4-pro (DeepSeek-V4-Pro-0813)gpt-6.2-astra, gpt-6.2-sol, gpt-6.1-terra, and gpt-6-luna are the four featured models; older GPT-6 and GPT-5 models stay listed
Newest releaseOctober 17, 2026 (V4.1-Flash-1017). V4-Pro-0813: August 13, 2026November 17, 2026 (GPT-6.2 Astra and GPT-6.2 Sol)
API priceFlash: $0.30 / $1.20 at peak, $0.15 / $0.60 off-peak. V4 Pro: $1.32 / $3.96 at peak, $0.66 / $1.98 off-peakGPT-6.2 Astra: $8 / $40. GPT-6.2 Sol: $2 / $10. GPT-6.1 Terra: $1 / $5. GPT-6 Luna: $0.10 / $0.50
Cached inputFlash: $0.006 at peak, $0.003 off-peakGPT-6.2 Sol: $0.10. GPT-6.2 Astra: $0.40
Context1M tokens, up to 384K out1,050,000 tokens (922,000 in, 128,000 out)
Images inFlash: yes. V4 Pro: "Not supported"Yes: "All latest OpenAI models support text and image input"
WeightsOpen, under the MIT license, on Hugging FaceClosed. OpenAI's open-weight models are gpt-oss-120b and gpt-oss-20b, under Apache 2.0, from August 2025
API formatOpenAI format and Anthropic format, plus the Responses APIResponses API and Chat Completions
Where data is stored"we directly collect, process and store your Personal Data in People's Republic of China"Servers "in the United States" and other countries; data residency projects in 10 regions
API data and trainingThe privacy policy lists training "our machine learning models and algorithms" as a use, with a right to opt out"data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us)"

DeepSeek's peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday. Its pricing page says "Off-peak rates are half of the peak rates", and weekends and Chinese public holidays are off-peak in full. OpenAI's prices above are for requests up to 272K input tokens; past that, "input and cache rates double and output costs 1.5 times the short-context rate for the full request."

DeepSeek vs OpenAI API pricing and token cost

Put the flagship coding models next to each other and DeepSeek costs a fraction per token. Put the smallest models next to each other and OpenAI is the cheaper one.

Price per million tokens, read on November 18, 2026.

ModelInputCached inputOutputBatch or off-peak
DeepSeek deepseek-flash (V4.1-Flash-1017)$0.30$0.006$1.20Off-peak: $0.15 / $0.60
DeepSeek deepseek-v4-pro (V4-Pro-0813)$1.32$0.044$3.96Off-peak: $0.66 / $1.98
OpenAI gpt-6-luna$0.10$0.01$0.50Batch: $0.05 / $0.25
OpenAI gpt-6.1-terra$1.00$0.05$5.00Batch: $0.50 / $2.50
OpenAI gpt-6.2-sol$2.00$0.10$10.00Batch: $1 / $5
OpenAI gpt-6.2-astra$8.00$0.40$40.00Batch: $4 / $20

What the table shows:

  • Against GPT-6.2 Sol, DeepSeek's Flash costs 6.7 times less for input and 8.3 times less for output at peak. Off-peak, the gap doubles.
  • Against GPT-6.2 Astra, Flash's output costs 33 times less at peak.
  • GPT-6 Luna undercuts both DeepSeek models: $0.10 / $0.50 against Flash's off-peak $0.15 / $0.60. OpenAI describes Luna as "our most efficient model for focused, high-volume tasks."
  • Discounts work differently. DeepSeek halves the price by the clock, on its own schedule. OpenAI halves it for Batch jobs, and its November 17 changelog adds background windows in the Responses API: "30% lower than Standard for the two-hour window and 40% lower for the eight-hour window."
  • Paying. DeepSeek deducts fees "from your topped-up balance or granted balance". Eligible ChatGPT Pro and Business subscribers "can now claim up to $500 per month in free API credits", per OpenAI's October 28 changelog.

Price per token isn't price per task: a model that takes more steps spends more tokens. The Terminal-Bench leaderboard in the next section prints the cost of each full run.

DeepSeek vs OpenAI benchmarks

Two sources put both vendors' models in one table: the Terminal-Bench leaderboard, run by the benchmark's own team, and DeepSeek's model card.

Terminal-Bench 4.1, from the leaderboard at tbench.ai, read on November 18, 2026. It scores coding agents on terminal tasks; the site says it is "Hosted by Stanford / Harbor / Laude Institute". Higher is better; ± is the 95% confidence interval.

Model (effort)AgentTasks resolvedCost of the run
GPT-6.1 Sol (max)Codex60.9% ± 3.0%$0.6k
GPT-6 Astra (max)Codex60.3% ± 2.8%$3.3k
GPT-6 Sol (max)Codex50.3% ± 3.2%$1.5k
GPT-5.6 Sol (max)Codex38.5% ± 3.8%$2.5k
DeepSeek V4.1 Flash 1017 (max)Claude Code38.5% ± 3.8%$0.2k
DeepSeek V4 Pro 0813 (max)Claude Code6.7% ± 2.5%$0.7k

GPT-6.2 Astra and GPT-6.2 Sol, one day old, aren't on the leaderboard yet. DeepSeek's Flash ties OpenAI's June flagship, GPT-5.6 Sol, at a twelfth of the run cost, and sits 22 points under GPT-6.1 Sol at a third of its cost.

DeepSeek's own table (the DeepSeek-V4.1-Flash-1017 model card) compares with GPT-5.6 Sol, the OpenAI model that was current before GPT-6:

Benchmark, as DeepSeek reports itDeepSeek-V4.1-Flash-1017GPT-5.6 Sol
DeepSWE v1.177.073.0
Terminal-Bench 4.042.339.9
HLE without tools42.644.5
ProgramBench29.123.0
Automation-Bench v1.0.660.545.8

DeepSeek's note: "We use the Claude Code harness for Terminal-Bench 4.0 and the Mini-SWE-V2 harness for DeepSWE v1.1." OpenAI's API docs for GPT-6.2 give prices, limits, and prompting advice, and no benchmark table. For which leaderboard measures what, see LLM benchmarks and leaderboards.

Is the DeepSeek API compatible with OpenAI's? Base URL and SDK

Yes. DeepSeek's docs say: "The DeepSeek API uses an API format compatible with OpenAI/Anthropic." You keep the OpenAI SDK and change two values:

from openai import OpenAI

client = OpenAI(
    api_key="<your DeepSeek API key>",
    base_url="https://api.deepseek.com",
)
  • Base URL (OpenAI format): https://api.deepseek.com
  • Base URL (Anthropic format): https://api.deepseek.com/anthropic
  • Model names: deepseek-flash or deepseek-v4-pro
  • Responses API: supported on both models. DeepSeek's guide: "To meet the demand for Codex, our API now supports the Responses API format". Fields that depend on stored state take fixed values, such as store: false and previous_response_id: null.
  • JSON output and tool calls: supported on both models.
  • Concurrency: 2,500 requests at once on Flash, 500 on V4 Pro.

Going the other way, OpenAI's GPT-6.2 models "support the Responses API and Chat Completions", and "Tool calling requires the Responses API."

Open weights and licence: DeepSeek is open, OpenAI's current models are closed

DeepSeek publishes the weights of both API models on Hugging Face. The model card says: "This repository and the model weights are licensed under the MIT License." You can run them on your own servers, fine-tune them, and sell what you build, keeping the copyright notice.

OpenAI's GPT-6 and GPT-6.2 models run only on OpenAI's API and the clouds it partners with. OpenAI's open-weight models are gpt-oss-120b ("Most powerful open-weight model, fits into an H100 GPU") and gpt-oss-20b, under a "Permissive Apache 2.0 license", with a 131,072-token context. Both went on Hugging Face in August 2025.

Outputs are yours on both. DeepSeek's Open Platform Terms: "We assign any rights, title, and interests—if any—in the Outputs of the Services to you", and you may use them for "training other models (such as model distillation)". OpenAI's Services Agreement: the customer "owns all Output."

Where you can use each one

DeepSeekOpenAI
Own APIapi.deepseek.com. DeepSeek's docs list no supported-country page208 countries and territories on OpenAI's supported list. China, Hong Kong, and Russia aren't on it
CloudsDeepSeek doesn't say on its own pages. Chinese AI models lists the clouds that carry itAmazon Bedrock "runs supported OpenAI models on AWS-managed infrastructure". Google Cloud is in private preview
OpenRouterDeepSeek V4.1 Flash 1017: 50 endpoints, and first in OpenRouter's weekly ranking by tokens processed (61.2T)GPT-6.2 Astra, on OpenAI's own endpoint at $8 / $40
Your own hardwareYes, the MIT weightsgpt-oss-120b and gpt-oss-20b only
Coding agentsSetup guides for Claude Code, Codex, OpenCode, OpenClaw, Hermes, and its own DeepSeek HarnessCodex. GPT-6.2 Astra supports hosted shell, apply patch, skills, computer use, and MCP tools in the Responses API

Data location and privacy, per each vendor's own policy

This section quotes the policies; it isn't legal advice.

DeepSeek. The API is run by Hangzhou DeepSeek Artificial Intelligence Co., Ltd. Its privacy policy (UK edition, last updated March 27, 2026) says: "To provide you with our services, we directly collect, process and store your Personal Data in People's Republic of China." It lists improving the service and training "our machine learning models and algorithms" among its uses, and gives you "the right to opt-out of using your Personal Data for training our models or optimizing our technologies." Since November 13, 2026, apps that serve the public through DeepSeek's built-in tools "must enable safety checks", and developers "remain responsible for content moderation, applicable model registration, and AI-generated content labeling."

OpenAI. Outside Europe the controller is OpenAI OpCo, LLC, in San Francisco; in the EEA and Switzerland it is OpenAI Ireland Limited. The privacy policy (updated February 6, 2026) says OpenAI processes data "in our facilities and servers in the United States" and in other countries where its partners and vendors are. For the API:

  • Training: "As of March 1, 2023, data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us)."
  • Retention: abuse monitoring logs are "retained for up to 30 days" by default. Approved customers get Zero Data Retention.
  • Data residency: eligible organizations create a project in a region and call its domain, such as eu.api.openai.com or jp.api.openai.com. Regional endpoints carry a 10% uplift on models released on or after March 5, 2026.

To keep prompts inside your own walls with DeepSeek, run its MIT weights yourself.

DeepSeek vs OpenAI vs Anthropic (Claude)

Anthropic sits at OpenAI's end of the price range and, like OpenAI, keeps its weights closed. The three-way comparisons live on their own pages:

Which to pick

If you wantPickWhy, from the vendors' own pages
The lowest bill for long agent runsDeepSeek deepseek-flash$0.30 / $1.20 at peak, half off-peak; $0.2k for the full Terminal-Bench 4.1 run
The highest score on coding-agent tasksOpenAI GPT-6.1 Sol or GPT-6.2 Sol, both $2 / $10GPT-6.1 Sol resolves 60.9% on Terminal-Bench 4.1; GPT-6.2 Sol is OpenAI's "go-to model for everyday work, from coding to powering agents at scale"
The cheapest tokens for simple, high-volume callsOpenAI GPT-6 Luna$0.10 / $0.50, and $0.05 / $0.25 in Batch
The model on your own serversDeepSeekMIT weights on Hugging Face
Prompts stored in the US, Europe, or JapanOpenAIData residency projects per region
No training on your API data by defaultOpenAINot used for training since March 1, 2023
One codebase for bothEitherDeepSeek speaks the OpenAI format: change base_url and the model name

You can run both from one codebase: DeepSeek for the bulk of agent steps and OpenAI for the hard ones. The SDK is the same, so the switch is one config value.

Publish what the model built

Ask either model for a landing page as one HTML file and you get a complete page back. Drop that file into Convika's publish tool, click Get my preview link, and you have a private link that works for 3 minutes, with no account. Publish it from a free account to keep it at one address, and Add a form to my page collects the entries people send.

Frequently asked questions

Is DeepSeek cheaper than OpenAI?

For the main coding models, yes. DeepSeek-V4.1-Flash-1017 costs $0.30 / $1.20 per million tokens at peak and $0.15 / $0.60 off-peak, against $2 / $10 for GPT-6.2 Sol and $8 / $40 for GPT-6.2 Astra. OpenAI's smallest model, GPT-6 Luna, costs $0.10 / $0.50, less than either DeepSeek model. Prices read on November 18, 2026.

Is DeepSeek better than OpenAI?

On the Terminal-Bench 4.1 leaderboard, OpenAI's GPT-6.1 Sol resolves 60.9% of tasks and DeepSeek V4.1 Flash 1017 resolves 38.5%, level with GPT-5.6 Sol. DeepSeek's own model card reports 77.0 on DeepSWE v1.1 against 73.0 for GPT-5.6 Sol. DeepSeek's run cost $0.2k on Terminal-Bench, GPT-6.1 Sol's $0.6k.

Is the DeepSeek API compatible with the OpenAI SDK?

Yes. DeepSeek's docs say its API "uses an API format compatible with OpenAI/Anthropic." Install the OpenAI SDK, set base_url to https://api.deepseek.com, use your DeepSeek key, and call deepseek-flash or deepseek-v4-pro. The Responses API format works too.

What is the DeepSeek OpenAI-compatible base URL?

https://api.deepseek.com for the OpenAI format, including Chat Completions and the Responses API. The Anthropic-format base URL is https://api.deepseek.com/anthropic.

Is DeepSeek open source? Is OpenAI?

DeepSeek's current models, DeepSeek-V4.1-Flash-1017 and DeepSeek-V4-Pro-0813, have open weights under the MIT license. OpenAI's GPT-6 and GPT-6.2 models are closed. OpenAI's open-weight models are gpt-oss-120b and gpt-oss-20b, under Apache 2.0, from August 2025.

Where does DeepSeek store data, and where does OpenAI?

DeepSeek's privacy policy says it stores personal data "in People's Republic of China." OpenAI's says it processes data on servers "in the United States" and other countries, and its API lets eligible organizations pin a project to one of 10 regions, including Europe and Japan.

Does OpenAI or DeepSeek train on API data?

OpenAI: "data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us)." DeepSeek's privacy policy lists training its models among the uses of personal data and gives a right to opt out.

What is OpenAI's newest model, and DeepSeek's?

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, 2026. DeepSeek's newest is DeepSeek-V4.1-Flash-1017, from October 17, 2026, served as deepseek-flash; its larger model is DeepSeek-V4-Pro-0813.

Sources

Each page was read on November 18, 2026, between 03:30 and 03:45 Japan time; dates in brackets are the page's own.

Next step

How to password protect an HTML page or website

A JavaScript password script leaves the password in the page source. To password protect an HTML page or link, encrypt the file or let a server check it.

Related articles

More ways to send your page

Skipbox

Cheap & fast GitHub Actions runners, unlimited for $39 a month

Your own machine keeps Docker layers and package caches between jobs. Switch by changing one runs-on line.

Start a free Skipbox trial⁠

Share this article

Publish the page your model wrote

Drop in the HTML file DeepSeek or GPT made. Convika puts it at a lasting link, keeps the form entries, and shows who visited.

Publish my HTML page free

Free plan: no credit card, 1 published page, 50 form entries a month.