Is DeepSeek V3.2 free?
DeepSeek advertises free chat in its app, but that does not guarantee you can choose DeepSeek V3.2 there. DeepSeek describes the app as powered by its latest flagship model; it does not promise that V3.2 is selectable. For production API use, V3.2 is paid on the OpenRouter provider routes listed for it. DeepSeek’s API pricing page does not list a current V3.2 price, so it does not give a current official quote for that model. [1][3][4]
DeepSeek V3.2 and V3.2-Exp are distinct names: DeepSeek describes V3.2 as the official successor to V3.2-Exp. The prices below are for V3.2 through OpenRouter, not official DeepSeek API rates or older V3.2-Exp prices. [2][4]
| Access route | Is it free? | What to know |
|---|---|---|
| DeepSeek app | Free chat is advertised | DeepSeek describes the app as powered by its latest flagship model; that does not establish that you can select or are using V3.2. [1] |
| DeepSeek API | No current V3.2 price confirmed in DeepSeek’s pricing table | The table lists other models, not V3.2, so don’t treat historical V3.2 API prices as a current official quote. [3] |
| OpenRouter API | Paid for the listed V3.2 provider routes | OpenRouter lists V3.2 with provider-specific prices. Its free model options do not make V3.2 free; free access is limited to models with a free variant. [4][5] |
| Self-hosting | The model weights are listed under an MIT license | The license does not pay for the compute, hardware, power, or hosting needed to run the model. [6] |
DeepSeek V3.2 API pricing and a sample budget
OpenRouter’s V3.2 listing identifies GMICloud as the provider for the displayed discounted rates: $0.2088 per 1 million input tokens and $0.3096 per 1 million output tokens, labeled 28% off. The same listing shows higher rates for other providers. To target GMICloud rather than rely on another route, use OpenRouter’s provider-routing controls to pin or select that provider; routing to another provider can change the price. The listing shows the discount but does not specify its eligibility or duration. [4]
Example at the displayed GMICloud rates:
- 100,000 input tokens: 0.1 × $0.2088 = $0.02088
- 20,000 output tokens: 0.02 × $0.3096 = $0.006192
- Estimated model usage total: $0.027072, or about 2.7 cents
For 1 million input plus 1 million output tokens, the same rates give $0.5184 in model usage. These figures are inference costs only—not the amount you must initially fund. OpenRouter’s pricing information says credit-purchase fees are separate from inference: Standard accounts pay 5.5% when buying credits, and Business accounts pay 8%; the fee is not charged on each request. OpenRouter’s terms set a $5 minimum credit purchase. So a $5 purchase corresponds to a $0.275 fee on Standard or $0.40 on Business at those percentages; the cited fee schedule does not state a separate minimum fee. [7][8]
OpenRouter says it passes through provider inference prices without a markup. Your total usage cost still depends on the provider route selected, its rates, token counts, and the separate fee when you buy credits. [7] The model page’s displayed discount does not state how long it lasts or any eligibility conditions, so confirm the route and price shown in your account before budgeting against it. [4]
Is V3.2 available and suitable for production API use?
DeepSeek’s release announcement says V3.2 launched on its app, web, and API. However, DeepSeek’s API pricing table does not list V3.2, while OpenRouter lists it through multiple providers. That means the sources support availability through the third-party OpenRouter route, but do not establish a current, directly priced official DeepSeek API offering for V3.2. [2][3][4]
OpenRouter’s V3.2 listing specifies a 164K context window and lists reasoning, tool calling, and structured outputs as model capabilities. These are model-page specifications, not a guarantee that every provider route offers the same limits or service. Confirm the selected provider’s current limits, availability, and data-handling terms before committing a production workload. [4]
DeepSeek’s official Hugging Face model repository lists the V3.2 weights under the MIT License. That can make self-hosting an option, but it does not make inference cost-free: infrastructure and operations still have costs. Bottom line: use the official app if free chat with DeepSeek’s current flagship is enough; if you specifically need V3.2 for production API calls, budget for a paid provider route and verify the route’s live price rather than assuming app access or open weights mean free API usage. [1][6]
Sources
DeepSeek V3.2 — OpenRouter pricing, providers, and specifications