Short answer
The full DeepSeek R1-0528 is listed as a paid OpenRouter API model, at $0.50 per million input tokens and $2.15 per million output tokens. OpenRouter also lists a free :free Qwen3-8B distilled variant, but that is a smaller, different model—not free API access to the full R1-0528. DeepSeek’s full-model repository is marked MIT-licensed, which permits use of the software under that license; it does not make hosted inference, hardware, or electricity free. (openrouter.ai)
OpenRouter’s current documentation says free variants have no model-use charge, but may have different rate limits or availability than paid models. Its documentation says daily free-request limits depend on credits purchased and reset at UTC midnight. I could not confirm from the current documentation retrieved here whether a new account can use this particular model through the API without first buying credits, the exact current per-minute or daily caps, or whether a negative balance blocks this model. Check the account’s live quota before relying on it. (openrouter.ai)
DeepSeek R1-0528 access and cost at a glance
| Route | Model or artifact | Price or cost | What to know |
|---|---|---|---|
| OpenRouter API: full model | deepseek/deepseek-r1-0528 |
$0.50/M input; $2.15/M output tokens | Paid API listing. Provider and routing choices can affect the rate. |
| OpenRouter API: free distill | deepseek/deepseek-r1-0528-qwen3-8b:free |
$0 per token on the free variant | An 8B distilled model, not the full model. Free-model limits and availability apply; current account-specific caps were not confirmed here. |
| Run full-model weights locally | DeepSeek R1-0528 repository, marked MIT | No per-token API bill for local inference | Storage, suitable hardware or rented compute, and electricity still have costs. |
| DeepSeek’s own API | Current official pricing page | No current R1-0528 API rate verified | Do not substitute prices for other DeepSeek models. |
OpenRouter’s full-model page lists provider rates of $0.500/$2.150 per million input/output tokens for DeepInfra, $0.500/$2.180 for SiliconFlow, and $0.571/$2.286 for StreamLake. These are listed provider rates, not a guarantee of which provider will serve a request; check the model page and routing settings before budgeting. (openrouter.ai)
DeepSeek’s official pricing page currently lists other models, not R1-0528. That means this article does not establish a direct DeepSeek API price for R1-0528; the paid rate above is OpenRouter’s listing.
Example API budget for the full model
At OpenRouter’s listed $0.50/M input and $2.15/M output rates, a workload of 1 million input tokens plus 250,000 output tokens comes to:
- Input: 1 × $0.50 = $0.50
- Output: 0.25 × $2.15 = $0.5375
- Estimated total: $1.0375
This is a calculation from the listed token rates, not a usage test or quote. It excludes any credit-purchase fees and assumes those rates apply to your selected provider and requests. (openrouter.ai)
Is the free OpenRouter model the same as full R1-0528?
No. The free listing identified in the available research is DeepSeek R1 0528 Qwen3 8B, with model ID deepseek/deepseek-r1-0528-qwen3-8b:free. It is a distilled 8B model. The full R1-0528 listing is a paid API model. Check the listing before use, since model availability and free endpoints can change. (openrouter.ai)
OpenRouter’s free-variant documentation says a :free variant is a separate catalog entry and may have different rate limits or availability from its paid counterpart. Its current daily-quota documentation says the limit depends on total credits purchased, and the counter resets at UTC midnight. OpenRouter provides an account-key endpoint for checking the quota that applies to your account. The documentation retrieved for this article did not establish the exact current thresholds, current per-minute cap, whether an initial credit purchase is required for this specific free model, or whether a negative balance blocks its requests. Treat those details as unconfirmed, rather than relying on older quoted limits. (openrouter.ai)
Watch for paid fallbacks: OpenRouter says fallback models are billed at their own rates. If you want to keep requests free, configure only free models in the fallback list, or use the free-only router; otherwise a paid fallback can create charges when the free model is unavailable or rate-limited. (openrouter.ai)
OpenRouter describes free models as suitable for experimentation, learning, and low-volume use, and says paid models are worth considering for production workloads with higher reliability requirements. A free listing alone does not confirm that a particular endpoint meets your production needs. (openrouter.ai)
Can I use the full model locally for free?
DeepSeek’s Hugging Face repository marks the full R1-0528 model MIT-licensed and lists the repository at about 689 GB. That supports access to the model files under the stated license, but local inference still requires substantial storage and computing resources. Unsloth describes a separate quantized full-model GGUF at roughly 185 GB and recommends at least 64 GB RAM for the quant it discusses. These are different artifacts; check the exact download before planning storage or hardware. (huggingface.co)
A smaller local option is the Qwen3-8B distill. LM Studio’s listing for that specific model reports a 4 GB minimum system-memory figure. Do not apply that figure to the full R1-0528. The listing itself does not establish price or license terms.
Which route makes sense?
- For a no-token-charge test: the Qwen3-8B
:freelisting is the free API option identified here. It is not the full model, and the account-specific quota and availability conditions should be checked before use. - For the full R1-0528 through an API: OpenRouter lists paid access. Use the live model page and selected provider’s rates for a budget; do not assume DeepSeek’s own API offers this exact model.
- For local control: the MIT-marked full-model weights are a possible local route, but compute, storage, power, or rented infrastructure are not free. The 8B distill is a different, smaller model.
Bottom line: “Free” can mean free model weights, a limited free hosted endpoint, or $0 token pricing for a particular API variant. Those are not interchangeable. The full R1-0528 has a listed paid OpenRouter API rate; the free Qwen3-8B distill is a separate model, and its current account limits should be confirmed before depending on it.
Sources
R1 0528 - API Pricing & Benchmarks | OpenRouter
DeepSeek R1 0528 Qwen3 8B free listing | OpenRouter
Free Variant - Free Model Access | OpenRouter
FreeModelDailyRequests | OpenRouter Documentation
API Credit & Rate Limits | OpenRouter
Free Models Router | OpenRouter
Models & Pricing | DeepSeek API Docs
DeepSeek-R1-0528 model repository | Hugging Face
DeepSeek-R1-0528 license | Hugging Face