Kimi K2.6

Open-weights agentic model tuned for tool use. Long, stable tool registries make it a good reuse case under automatic caching.

$0.95
Input / 1M tokens
$4.00
Output / 1M tokens
$0.16
Cache read · −83%
262K
Context window

At a glance.

ProviderFireworks AI
Familykimi_k25
Released2026-04
LicenseOpen weights
Context window262K tokens
Max output262K tokens
Parameters1029B
Modalitiestext, image
Tool callingYes
Reasoning modeYes
Cachingautomatic
Batch discountNo batch tier

What reuse looks like here.

Kimi K2.6 · agent trafficper request
Total input100%
Candidate reuse62%
Realized reuse50%
Capture rate81%
140ms
Warm TTFT · −63% vs cold
250
Output tokens / sec
Reuse economics

What you actually pay once caching works.

At a typical 55% prefix reuse, a million input tokens on Kimi K2.6 effectively costs $0.52 instead of $0.95 - blending to roughly $1.39 with a 25% output share. There is no batch tier, so cost control here leans on caching and routing.

Estimate it for your workload
Best for
agentictool-usecoding

Routes through these aliases:

Same OpenAI client, this model.

python
from openai import OpenAI

client = OpenAI(base_url="https://api.zumik.ai/v1", api_key="zk_live_...")

r = client.responses.create(
    model="kimi-k2p6",          # or an alias like code.balanced
    input="Draft a fix for the failing test.",
)
print(r.usage.input_tokens_cached)   # confirm reuse

Kimi K2.6, answered.

How much does Kimi K2.6 cost?

Kimi K2.6 is $0.95 per million input tokens and $4.00 per million output tokens through Zumik. Cache reads are $0.16 per million, a 83% discount on input.

What is Kimi K2.6's context window?

Kimi K2.6 supports a 262K-token context window with up to 262K output tokens.

Does Kimi K2.6 support prompt caching?

Yes. Fireworks AI uses Automatic prompt caching (serverless and dedicated) caching. In the Zumik corpus, Kimi K2.6 shows a median cache capture of 81% on agent workloads.

Which Zumik aliases route to Kimi K2.6?

Kimi K2.6 is a candidate for the code.balanced, auto.cheapest aliases, selected when it wins under current routing policy.

Run Kimi K2.6 with reuse measured.

Point an OpenAI client at Zumik and see exactly how much of this model's input you are reusing.