Kimi K3 API pricing

21 providers serve Kimi K3. Kimi K3 is an open-weight model (Kimi K3) with 2800B total parameters, 104B active per token (mixture-of-experts), up to 1049K context. Prices below are per 1M tokens, cheapest input first.

Refreshed 2026-11-22 · list prices from provider APIs
Input floor $0.192/1M at OpenRouter Output floor $12.50/1M at Sail and Aiand Find your workload winner →
Provider Input $/1M Output $/1M Cache read $/1M Context
OpenRouter marketplace quote $0.192 $16.00 $0.182 1049K Use
Sail $2.50 $12.50output floor $0.250 1049K
Scx AI $2.83 $14.13 $0.280 1000K
DeepInfra $2.85 $14.25 $0.285 1049K Use
Databricks $3.00 $15.00 $0.300 1000K Use
Github Copilot $3.00 $15.00 $0.300 918K
Moonshot $3.00 $15.00 $0.300 1049K Use
Nebius $3.00 $15.00 $0.300 1049K Use
Together AI $3.00 $15.00 $0.300 1049K Use
Tencent $3.00 $15.00 $0.300 1000K
Fireworks AI $3.00 $15.00 $0.300 1049K Use
Novita AI $3.00 $15.00 $0.300 1049K Use
Aihubmix $3.00 $15.00 $0.300 1049K
Aiand $3.00 $12.50 $0.500 1049K
Baseten $3.00 $15.00 $0.300 1049K Use
Opencode Go $3.00 $15.00 $0.300 1049K
Empiriolabs $3.00 $15.00 - 1049K
Wally $3.00 $15.00 $0.300 1049K
GMI Cloud $3.00 $15.00 $0.300 1049K Use
Kyma $3.05 $15.26 $0.405 1049K
AWS Bedrock $3.30 $16.50 $0.330 1000K Use

The cheapest way to run Kimi K3

On an illustrative 70M-input / 30M-output monthly workload, today's lowest listed cost is roughly $493/month, at OpenRouter. Input-heavy, output-heavy and cache-heavy workloads can produce different winners. Use the live mix calculator below rather than combining floors from two different providers. On output-token cost alone, the API floor remains below our self-hosting estimate of ~$14.96/1M. Details below. Run your own numbers in the breakeven calculator, or read the breakeven math.

Your actual monthly cost

Uses today's list prices

Cache read uses the listed cache rate where available; otherwise it falls back to normal input price. Batch, write-cache, volume and negotiated discounts are excluded.

Or run it yourself: Kimi K3 self-hosting economics

Kimi K3 is open-weight (Kimi K3), so the API price above competes with the GPU-hour market. 2800B total parameters (MoE, ~104B active per token) needs roughly 3851 GB VRAM at FP8 or 1926 GB at INT4, KV-cache headroom included.

No single GPU fits Kimi K3 at FP8 (3851 GB needed), so it needs a multi-GPU node. Cheapest tracked option: 21× MI300X (192 GB each) at roughly $10.50/hr total (RunPod). Rent 21× MI300X at RunPod → INT4 quantization drops the requirement to 1926 GB. Tune it in the calculator.
Breakeven estimate: a well-batched vLLM deployment on 21× MI300X ($10.50/hr) at ~390 aggregate tok/s and 50% utilization works out to roughly $14.96 per 1M output tokens, versus $12.50 via the cheapest output-token API (Sail and Aiand). The API is cheaper unless your utilization or throughput beats these assumptions. Planning estimate: throughput varies with hardware, quantization, batch size and context. Tune it in the calculator.

Related models

ModelCheapest in $/1MCheapest out $/1MProviders
Kimi K2.7 Code $0.380 $1.60 23
Kimi K2.6 $0.380 $1.60 21
Kimi K3.1 Pro $0.240 $8.40 13
Kimi K2.5 $0.450 $2.25 11
Kimi K2 Thinking $0.600 $1.20 10
Kimi K2 $0.500 $2.00 10

FAQ

What is the cheapest API for Kimi K3?

As of 2026-11-22, the lowest input price for Kimi K3 is OpenRouter at $0.192 per 1M input tokens. The lowest output price, $12.50 per 1M tokens, is likewise shared by Sail and Aiand. The cheapest provider for a real workload depends on its input/output mix. The next-lowest input price is Sail at $2.50, a 1202% difference.

How much VRAM do you need to self-host Kimi K3?

Kimi K3 has 2800B parameters, so plan for roughly 3851 GB of VRAM at FP8 or 1926 GB at INT4/AWQ, KV-cache headroom included. That exceeds a single GPU. A typical node is 21× MI300X.

← All models · GPU rental prices · Breakeven calculator · Get this data as JSON