Short answer: DeepSeek advertises free chat with its latest flagship model, but that offer does not promise free access to DeepSeek-V3.1 or free API calls. DeepSeek’s current API documentation lists newer models. Its legacy deepseek-chat and deepseek-reasoner names became inaccessible after July 24, 2026. You can still find V3.1 weights, but downloading weights does not pay for the hardware or compute needed to run them. (DeepSeek)
Is DeepSeek V3.1 free?
It depends on the access route:
- Chat: DeepSeek’s official page advertises free chat with its latest flagship model. It does not promise that the chat runs V3.1 specifically or specify a V3.1 free-use allowance. (DeepSeek)
- Official hosted API: The current API model names are
deepseek-flashanddeepseek-v4-pro, not V3.1. Do not budget on V3.1 being available as a free official API model. (DeepSeek API pricing) - V3.1 weights: The Replicate-hosted DeepSeek-V3.1 model page states that the repository and model weights are licensed under the MIT License, and links to download the DeepSeek-V3.1 weights on Hugging Face. That license statement applies to the repository and weights; it does not make hosted inference, hardware, or compute free. You must arrange and pay for any hardware or cloud compute needed to run the model.
Third-party pages advertising “free chat” or gateway credits do not establish official API terms. Treat introductory credits as temporary promotions—not an ongoing free tier—unless the provider’s current terms specify the amount, expiry, eligible models, and limits.
What should you budget for the official API?
DeepSeek’s current pricing table does not list V3.1. For the currently documented deepseek-flash API, it lists prices per 1 million tokens, with separate rates for cache-hit input, cache-miss input, and output. The table displays a $ symbol but does not specify a currency code. (DeepSeek API pricing)
| Current API model | Input, cache miss (off-peak / peak) | Input, cache hit (off-peak / peak) | Output (off-peak / peak) |
|---|---|---|---|
deepseek-flash (served by V4.1-Flash-1017) |
$0.15 / $0.30 per 1M tokens | $0.003 / $0.006 per 1M tokens | $0.60 / $1.20 per 1M tokens |
The pricing page defines peak hours as 01:00–04:00 and 06:00–10:00 UTC, Monday through Friday, excluding Chinese public holidays. All other hours—including weekends and Chinese public holidays—are off-peak. DeepSeek says prices can change, so confirm the current table before forecasting spend. The rates above are those displayed when checked on November 10, 2026. (DeepSeek API pricing)
Example budget
Assume monthly usage of 1 million uncached input tokens and 100,000 output tokens, with no Search or Fetch requests:
- All off-peak: (1 × $0.15) + (0.1 × $0.60) = $0.21
- All peak: (1 × $0.30) + (0.1 × $1.20) = $0.42
These are examples using the listed deepseek-flash rates—not V3.1 prices or a forecast of your bill. Your actual cost depends on token usage, cache status, and when requests are processed. The dollar symbol is shown as published; the pricing page does not identify a currency code. (DeepSeek API pricing)
DeepSeek separately lists Search and Fetch at $0.75 per 1,000 successful requests for each API. Its pricing page also says server-side Search and Fetch calls through DeepSeek Harness, Claude Code, or Codex became billable on November 1, 2026. Check the page’s billing rules for the route you use. (DeepSeek API pricing)
To estimate your own spend, count input and output tokens separately, apply the selected model’s rates and the applicable peak/off-peak schedule, and add any separately billed tools. Keep temporary gateway credits separate from recurring paid usage.
Which route fits your use?
| What you want | Practical route | Budget implication |
|---|---|---|
| Try DeepSeek casually | Use the official app or website | DeepSeek advertises free chat with its latest flagship model; this is not a promise that the chat specifically uses V3.1. (DeepSeek) |
| Build against DeepSeek’s current hosted API | Use the current API documentation and a supported model ID, such as deepseek-flash |
API usage is priced by tokens under the current price table. The API guide says to obtain an API key. (DeepSeek API pricing) |
| Run V3.1 specifically | Download the DeepSeek-V3.1 weights and arrange your own inference setup, or find a provider that explicitly lists V3.1 | The Replicate-hosted model page states that the repository and weights are MIT-licensed. This is a licensing statement, not a promise of free hosting or compute; separately account for hardware, deployment, and operations. (Replicate model page) |
For the current API, DeepSeek documents the OpenAI-compatible base URL as https://api.deepseek.com, requires an API key, and shows Bearer authentication in its example. Its current model table lists deepseek-flash and deepseek-v4-pro; it gives both a 1M-token context length and says both support thinking and non-thinking modes. These are current API specifications, not V3.1 specifications. (DeepSeek API example)
What V3.1 could do—and what matters now
DeepSeek’s V3.1 release announcement described thinking and non-thinking modes, 128K context for the API models at that time, and beta strict function calling. Those are historical release details; they do not establish what a provider serves today. (DeepSeek V3.1 release announcement)
Before a production integration, verify the selected provider’s current model ID, context limit, mode and tool support, pricing, and service limits. The key question is not only whether V3.1 weights can be downloaded, but whether your chosen provider serves that exact model and what it costs to operate.
Bottom line: DeepSeek advertises free chat with its latest model, but that does not establish free V3.1 API access. The official API has moved on from the V3.1-era model names. For production budgeting, use the current rate table for a supported model—or separately estimate the compute and operating costs of self-hosting V3.1.
Sources
Models & Pricing | DeepSeek API Docs
Your First API Call | DeepSeek API Docs