DeepSeek R1 API providers: 0528 costs

Choose DeepInfra via OpenRouter for the lowest listed example cost: $7.54325 for 1,000 requests.

Updated · 5 sources · Methodology

NeedPickWhy
Lowest example costDeepInfra via OpenRouter$7.5432514
Fastest listed routed outputStreamLake via OpenRouterOpenRouter P50 records 43 tokens/s for this route.5
Shortest listed routed initial waitDeepInfra via OpenRouterOpenRouter P50 records an initial wait of 0.63 s for this route.5

Performance

ProviderAPI and model IDPublished route P50 output (tokens/s)Published route P50 initial response (s)
DeepInfra via OpenRouter5DeepSeek R1 0528320.63
SiliconFlow via OpenRouter5DeepSeek R1 0528191.83
StreamLake via OpenRouter5DeepSeek R1 0528433.42
ProviderServing path
DeepInfraOpenRouter route
SiliconFlowOpenRouter route
StreamLakeOpenRouter route
Novita AIDirect provider API
Vertex AIDirect provider API

Prices and request cost

ProviderCost / 10K input + 1K output request1,000-request cost
DeepInfra via OpenRouter14$0.00754325$7.54325
SiliconFlow via OpenRouter14$0.0075749$7.5749
StreamLake via OpenRouter14$0.00843578$8.43578
Novita AI direct2$0.0095$9.5
Vertex AI direct3$0.0189$18.9
API documentationNovita AI · Chat API
Cost example: 1,000 requests, each 10,000 input + 1,000 billed output tokens
DeepInfra via OpenRouter$7.54325
SiliconFlow via OpenRouter$7.5749
StreamLake via OpenRouter$8.43578
Novita AI direct$9.5
Vertex AI direct$18.9

Options, strengths and limitations

  1. DeepInfra via OpenRouter

    DeepInfra supplies the lowest-cost R1 0528 route in this OpenRouter shortlist.

    Cost / 10K input + 1K output request
    $0.00754325
    Example quantity
    1000
    Example total
    $7.54325

    Strengths

    • OpenRouter P50 records an initial wait of 0.63 s for this route.1

    Limitations

    • DeepInfra’s routed output is 32 tokens/s, below StreamLake’s 43 tokens/s in the same P50 table.1

    Sources:1

  2. SiliconFlow via OpenRouter

    SiliconFlow is another upstream R1 0528 provider under the same OpenRouter account and routing interface.

    Cost / 10K input + 1K output request
    $0.0075749
    Example quantity
    1000
    Example total
    $7.5749

    Strengths

    • SiliconFlow matches DeepInfra’s $0.5 uncached input rate per 1M tokens; their output tariffs differ.1

    Limitations

    • At 19 tokens/s, SiliconFlow has the lowest output rate of the 3 displayed OpenRouter routes.1

    Sources:1

  3. StreamLake via OpenRouter

    StreamLake is the throughput-focused R1 0528 option among the displayed OpenRouter routes.

    Cost / 10K input + 1K output request
    $0.00843578
    Example quantity
    1000
    Example total
    $8.43578

    Strengths

    • OpenRouter P50 records 43 tokens/s for this route.1

    Limitations

    • StreamLake’s 3.42 s initial wait is longer than DeepInfra’s 0.63 s in this routed snapshot.1

    Sources:1

  4. Novita AI direct

    Novita’s direct R1 0528 API bypasses the OpenRouter routing account.

    Cost / 10K input + 1K output request
    $0.0095
    Example quantity
    1000
    Example total
    $9.5

    Strengths

    • A direct Novita request selects R1 0528 without relying on an OpenRouter upstream-provider choice.2

    Limitations

    • Novita’s direct example costs $9.5 per 1,000 requests, above the DeepInfra route’s $7.54325.21

    Sources:2

  5. Vertex AI direct

    Vertex AI provides R1 0528 through a Google-managed model API and Google Cloud billing.

    Cost / 10K input + 1K output request
    $0.0189
    Example quantity
    1000
    Example total
    $18.9

    Strengths

    • The Vertex AI entry keeps R1 0528 usage on a Google Cloud account rather than a separate router account.3

    Limitations

    • Vertex’s $18.9 example is the highest of these R1 0528 options, including the other direct API.321

    Sources:3

Methodology and sources

DeepSeek R1 0528: 1,000 request(s), each with 10,000 uncached input and 1,000 billed output tokens. Official seller tariffs and the named routing path define each price. OpenRouter P50 figures describe the listed routes; they are not direct-provider tests or a controlled shared-prompt benchmark. OpenRouter Standard charges $5.5 when $100 of credits are purchased; allocating that fee over fully consumed credits multiplies routed inference cost by 1.055. Business uses 1.08. The example excludes optional BYOK, tax and unspent balances. These route figures are the public OpenRouter standard text-generation P50 over a 30-minute window. Initial response is not the first final-answer token; requests do not share a controlled prompt or reasoning budget. Latency is converted from milliseconds to seconds. Rolling third-party performance values are the batch snapshot identified by the page’s Updated line, not live guarantees.5

  1. OpenRouter · DeepSeek R1 0528 specifications and pricing
  2. Novita AI direct · DeepSeek R1 0528 specifications and pricing
  3. Vertex AI direct · DeepSeek R1 0528 specifications and pricing
  4. OpenRouter · Standard and Business credit-purchase fees
  5. OpenRouter · public standard-route P50 statistics

Build with Novita Model APIs

Create an API key and connect your application.

Start BuildingExplore Models