Kimi K2 Thinking API pricing

9 providers serve Kimi K2 Thinking. Kimi K2 Thinking is an open-weight model (Modified MIT) with 1058B total parameters, 32B active per token (mixture-of-experts), up to 262K context. Prices below are per 1M tokens, cheapest input first (self-hosting math further down).

Refreshed 2026-08-23 · list prices from provider APIs

Input floor $0.600/1M at AWS Bedrock, Fireworks AI, Baseten and 4 more Output floor $1.20/1M at GMI Cloud Find your workload winner →

Provider	Input $/1M	Output $/1M	Cache read $/1M	Context
AWS Bedrock	$0.600	$2.50	-	128K	Use
Fireworks AI	$0.600	$2.50	-	262K	Use
Baseten	$0.600	$2.50	-	-	Use
Moonshot	$0.600	$2.50	$0.150	262K	Use
Vertex AI Moonshot Models	$0.600	$2.50	-	256K
Novita AI	$0.600	$2.50	-	262K	Use
OpenRouter marketplace quote	$0.600	$2.50	$0.150	262K	Use
GMI Cloud	$0.800	$1.20output floor	-	262K	Use
Crusoe	$2.50	$2.50	-	262K	Use

The cheapest way to run Kimi K2 Thinking

On an illustrative 70M-input / 30M-output monthly workload, today's lowest listed cost is roughly $92.00/month, at GMI Cloud. 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 ~$1.76/1M. Details below. Run your own numbers in the breakeven calculator, or read the breakeven math.

Your actual monthly cost

Uses today's list prices

Workload preset Input tokens / month (millions) Output tokens / month (millions) Input served from cache %

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 K2 Thinking self-hosting economics

Kimi K2 Thinking is open-weight (Modified MIT), so the API price above competes with the GPU-hour market. 1058B total parameters (MoE, ~32B active per token) needs roughly 1455 GB VRAM at FP8 or 728 GB at INT4, KV-cache headroom included.

No single GPU fits Kimi K2 Thinking at FP8 (1455 GB needed), so it needs a multi-GPU node. Cheapest tracked option: 8× MI300X (192 GB each) at roughly $4.00/hr total (RunPod). Rent 8× MI300X at RunPod → INT4 quantization drops the requirement to 728 GB. Tune it in the calculator.

Breakeven estimate: a well-batched vLLM deployment on 8× MI300X ($4.00/hr) at ~1265 aggregate tok/s and 50% utilization works out to roughly $1.76 per 1M output tokens, versus $1.20 via the cheapest output-token API (GMI Cloud). 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

Model	Cheapest in $/1M	Cheapest out $/1M	Providers
Kimi K2	$0.500	$2.00	11
Kimi K2.5	$0.500	$2.80	8
Kimi K2.6	$0.950	$4.00	6
Kimi K2.7 Code	$0.670	$3.40	4
Kimi K3	$3.00	$15.00	3
MoonshotAI Kimi Latest	$2.00	$5.00	2

FAQ

What is the cheapest API for Kimi K2 Thinking?

As of 2026-08-23, the lowest input price for Kimi K2 Thinking, $0.600 per 1M tokens, is the same list price at AWS Bedrock, Fireworks AI, Baseten and 4 more. The lowest output price is GMI Cloud at $1.20 per 1M output tokens. The cheapest provider for a real workload depends on its input/output mix. The next-lowest input price is GMI Cloud at $0.800, a 33% difference.

How much VRAM do you need to self-host Kimi K2 Thinking?

Kimi K2 Thinking has 1058B parameters, so plan for roughly 1455 GB of VRAM at FP8 or 728 GB at INT4/AWQ, KV-cache headroom included. That exceeds a single GPU. A typical node is 8× MI300X.

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