Best Kimi K2.5 API providers on OpenRouter

Choose AtlasCloud for the $0.007807 request example; Amazon Bedrock is the fastest of the 5 listed OpenRouter Kimi K2.5 routes at 50.5 tokens/s.

Updated · 3 sources · Methodology

NeedPickWhy
Lowest example costAtlasCloud$0.00780712
Highest published route P50 outputAmazon Bedrock50.5 tokens/s3
Lowest published route P50 initial waitVenice0.99 s3

Performance

ProviderPublished route P50 output (tokens/s)Published route P50 initial response (s)
AtlasCloud3223.02
SiliconFlow3231.58
Novita AI · OpenRouter3244.00
Amazon Bedrock350.51.21
Venice3450.99

Prices and request cost

ProviderInput / 1M tokensOutput / 1M tokens
AtlasCloud1$0.49$2.5
SiliconFlow1$0.5$2.5
Novita AI · OpenRouter1$0.6$3
Amazon Bedrock1$0.6$3
Venice1$0.56$3.5
API documentationNovita AI · Chat API
Cost example: 10,000 input + 1,000 billed output tokens
AtlasCloud$0.007807
SiliconFlow$0.0079125
Novita AI · OpenRouter$0.009495
Amazon Bedrock$0.009495
Venice$0.0096005

Options, strengths and limitations

  1. AtlasCloud

    AtlasCloud: Kimi K2.5 at 22 tokens/s in the OpenRouter P50 snapshot.

    Input / 1M tokens
    $0.49
    Output / 1M tokens
    $2.5
    Output speed (tokens/s)
    22

    Strengths

    • Kimi K2.5 costs $0.007807 per example on AtlasCloud; the measured initial wait is 3.02 s.11

    Limitations

    • At 22 tokens/s, AtlasCloud has the slowest output of these 5 routes.13

    Sources:11

  2. SiliconFlow

    SiliconFlow: Kimi K2.5 at 23 tokens/s in the OpenRouter P50 snapshot.

    Input / 1M tokens
    $0.5
    Output / 1M tokens
    $2.5
    Output speed (tokens/s)
    23

    Strengths

    • A 1.58 s initial wait accompanies the $0.0079125 Kimi K2.5 request example.11

    Limitations

    • SiliconFlow’s 23 tokens/s trails Amazon Bedrock’s 50.5 tokens/s in this route P50 table.13

    Sources:11

  3. Novita AI · OpenRouter

    Novita AI: Kimi K2.5 at 24 tokens/s in the OpenRouter P50 snapshot.

    Input / 1M tokens
    $0.6
    Output / 1M tokens
    $3
    Output speed (tokens/s)
    24

    Strengths

    • Novita AI pairs $0.009495 for the Kimi K2.5 token scenario with a 4.00 s initial wait.11

    Limitations

    • Novita AI’s 4.00 s initial response is the longest of the listed routes.13

    Sources:11

  4. Amazon Bedrock

    Amazon Bedrock: Kimi K2.5 at 50.5 tokens/s on the OpenRouter route.

    Input / 1M tokens
    $0.6
    Output / 1M tokens
    $3
    Output speed (tokens/s)
    50.5

    Strengths

    • The route combines 50.5 tokens/s output with a 1.21 s P50 initial wait.11

    Limitations

    • The request example costs $0.009495 versus AtlasCloud’s $0.007807; these are OpenRouter timings, not direct Bedrock measurements.1

    Sources:11

  5. Venice

    Venice: Kimi K2.5 at 45 tokens/s in the OpenRouter P50 snapshot.

    Input / 1M tokens
    $0.56
    Output / 1M tokens
    $3.5
    Output speed (tokens/s)
    45

    Strengths

    • At $0.0096005, the selected Kimi K2.5 request retains the route’s 0.99 s measured start.11

    Limitations

    • Venice’s $0.0096005 request example is the most expensive of these 5 routes.1

    Sources:11

Methodology and sources

Kimi K2.5: 1 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. All 5 entries are OpenRouter routes. Prices use standard tariffs before promotional discounts; direct-provider bills and limits may differ. Rolling third-party performance values are the batch snapshot identified by the page’s Updated line, not live guarantees.3

  1. OpenRouter · Kimi K2.5 route prices and P50 performance
  2. OpenRouter · Standard and Business credit-purchase fees
  3. OpenRouter · public standard-route P50 statistics

Build with Novita Model APIs

Create an API key and connect your application.

Start BuildingExplore Models