Skip to main content

Kimi K2.5

Live
Moonshot256K context · Text, Vision, Tools, ReasoningBenchmark price · $0.532 input / $3.33 output per 1M tokens
Use this model
Best discount now99.4%offYou pay 0.6% of reference.
Median discount99.2%offAcross 519 live offers.
Realized · 7 days82.7%avgRequest-weighted average.
Requests · 24h10,469519 offers from 57 sellers.
Current offers
518 of 692 offers
Availability
Providers
ProviderDiscountPriceLatencyUptimeCacheSold
Alibaba DashScope — China
99%
$0.00332 in
$0.0207 out
1.2s
96.5%
46.3%
$0.580
Amazon Bedrock
99%
$0.00411 in
$0.0205 out
916ms
85.0%
5.0%
$28
Alibaba DashScope — China
99%
$0.00411 in
$0.0205 out
1.2s
96.5%
46.3%
-
Alibaba DashScope — China
99%
$0.00411 in
$0.0205 out
1.2s
96.5%
46.3%
-
Alibaba DashScope — China
99%
$0.00411 in
$0.0205 out
1.2s
96.5%
46.3%
-
Alibaba DashScope — China
99%
$0.00411 in
$0.0205 out
1.2s
96.5%
46.3%
-
Alibaba DashScope — China
99%
$0.00411 in
$0.0205 out
1.2s
96.5%
46.3%
-
Alibaba DashScope — China
99%
$0.00411 in
$0.0205 out
1.2s
96.5%
46.3%
-
Alibaba DashScope — China
99%
$0.00411 in
$0.0205 out
1.2s
96.5%
46.3%
-
Alibaba DashScope — China
99%
$0.00411 in
$0.0205 out
1.2s
96.5%
46.3%
-
Total
Cumulative offers
Sold
$42

Discount trend

Realized · 7 days
99.6%+0.0 pts vs last week
Best, twice dailyMedian 99.2%

Recent fills

16:5340,878 tokensVenice AI48.3%
16:51170 tokensVenice AI46.6%
16:5040,281 tokensVenice AI48.0%
16:509 tokensAlibaba DashScope100.0%
16:4739,494 tokensVenice AI48.0%
16:4629,929 tokensVenice AI37.9%
16:45258 tokensAlibaba DashScope99.4%
16:4438,328 tokensVenice AI37.7%
16:4317,854 tokensVenice AI47.1%
16:4337,325 tokensVenice AI36.8%
16:42105 tokensAlibaba DashScope99.6%
16:422,011 tokensAWS Bedrock99.4%
16:42152 tokensVenice AI46.6%
16:409 tokensAlibaba DashScope100.0%
16:386,503 tokensVenice AI47.9%
16:3834 tokensAWS Bedrock100.0%
16:376,042 tokensVenice AI34.3%
16:376,563 tokensVenice AI38.8%
16:35258 tokensAlibaba DashScope99.4%
16:3514,241 tokensVenice AI40.3%
Times in UTCSettled on Base in USDC.

Performance by provider route · 24h

Measured on Surplus traffic
AWS Bedrock51% of traffic
85.0% success
p50 4.11 s · p95 24.87 s · 99 tok/s
Venice AI direct42% of traffic
99.6% success
p50 33.57 s · p95 111.50 s · 40 tok/s
Alibaba DashScope — China7% of traffic
96.5% success
p50 2.25 s · p95 10.11 s · 30 tok/s
Mordiem0% of traffic
46.1% success
p50 3.04 s · p95 - · 54 tok/s
Uncensored AI direct0% of traffic
82.6% success
p50 - · p95 - · -
InferHub0% of traffic
36.4% success
p50 - · p95 - · -
OpenRouter0% of traffic
1.0% success
p50 - · p95 - · -
Use this model

Change your base URL.
Keep your code.

Any OpenAI-compatible client works. Add /min30 to the path to only buy at 30% off or better.

Read the quickstart ↗
Parameters this model acceptsPOST /v1/chat/completions
  • modelstringThis model’s id
  • messagesarrayThe conversation, as role and content pairs
  • streambooleanSend the answer as server-sent events
  • max_tokensintegerMost output tokens to generate
  • temperaturenumberSampling temperature
  • top_pnumberNucleus sampling
  • top_kintegerTop-k sampling
  • min_pnumberDrops tokens below this share of the top token’s probability
  • stopstring | arrayStop sequences
  • seedintegerDeterministic sampling
  • toolsarrayFunctions the model may call, in the OpenAI format
  • tool_choicestring | objectauto, required, none, or one named function
  • parallel_tool_callsbooleanAllow several tool calls in one turn
  • response_formatobjectJSON mode or a JSON schema for structured output
  • structured_outputsbooleanAccepts a JSON schema in response_format
  • reasoningobjectReasoning options as an object, such as {"effort": "high"}
  • include_reasoningbooleanStripped before the request reaches the provider
  • frequency_penaltynumberPenalises tokens by how often they appeared
  • presence_penaltynumberPenalises tokens that appeared at all
  • repetition_penaltynumberPenalises repeated tokens
  • logit_biasobjectPer-token bias on the logits
  • logprobsbooleanReturn token log probabilities
  • top_logprobsintegerHow many top tokens to return probabilities for
Context window 256K tokens · Max output 66K tokensOthers are passed through; most providers ignore what they don’t support. Parameter compatibility ↗
from openai import OpenAI

client = OpenAI(
    api_key="inf_your_key_here",
    base_url="$SURPLUS_BASE_URL"
)

response = client.chat.completions.create(
    model="kimi-k2.5",
    messages=[{"role": "user",
               "content": "Hello!"}]
)

SURPLUS_BASE_URL
https://api.surplusintelligence.ai/v1

Related models