How much does Kimi K3 cost?

On the Respan gateway, input starts at $0.72 per 1M tokens on Relace, and output starts at $9.00 per 1M tokens on Wafer, after gateway discounts. Averaged across 17 providers, it costs $2.58 per 1M input tokens and $14.53 per 1M output tokens.

Which providers serve Kimi K3 on Respan?

17 providers serve Kimi K3: Relace, OpenRouter, Wafer, Makora, Novita, TogetherAI, and 10 more. TogetherAI, Fireworks, Perplexity, and Tera require your own provider API key (BYOK).

What is Kimi K3's context window?

Kimi K3 accepts up to 1.0M tokens of context on its largest provider route.

How do I use Kimi K3 with the Respan gateway?

Point any OpenAI-compatible SDK at https://api.respan.ai/api/ with your Respan API key and set the model to "moonshot/kimi-k3". To set it up with a coding agent, copy the agent setup prompt at the top of this page.

Over the last 30 days of requests through the Respan gateway, the fastest provider was Baseten at 74 tokens per second, and the lowest response latency was 1.6 seconds on Inference.net. Speed varies with request size and provider.

Where does Kimi K3 rank on the Respan Index?

Kimi K3 scores 50.6 out of 100 on the Respan Index v1.1, ranking #21 of 138 models. By category: Coding 37.2, Agents 63.1, Reasoning 47.0, Knowledge & language 52.7, and Human preference 56.7. The Respan Index combines independent and lab-reported benchmark results into one score, using benchmark data collected by llmmetric.com.

How does Kimi K3 score on benchmarks?

Kimi K3 has 85 published benchmark scores, including Terminal-Bench 2.1 88.3%, DeepSearchQA (F1) 95%, and GPQA Diamond 93.5%. Each score links to where it was published; the data is collected by llmmetric.com.

Kimi K3

Kimi K3 is a 2.8T parameter open-weight multimodal reasoning model from Moonshot AI. It is suited for complex coding, knowledge work, and long-horizon agentic workflows, and is particularly strong at navigating large repositories, using tools, debugging, and iterating against images, logs, tests, and runtime feedback. Its architecture uses KDA and Attention Residuals for computational efficiency.

Tool Use

# Paste into Claude Code, Cursor, or Codex

Help me set up the Respan gateway: fetch and follow https://respan.ai/docs/agent-setup.md. Use model "moonshot/kimi-k3".

Context window: 1.0M tokens
Input from: $0.72/M
Output from: $9.00/M
Providers: 17
Released: Jul 16, 2026

Providers

17 providers serve this model through the Respan gateway, cheapest first. Copy a model ID to call that route directly.

Provider	Context	Input	Output
Relace`relace/moonshotai/kimi-k3`	1.0M	$0.72/M	$20.00/M
OpenRouter`openrouter/moonshotai/kimi-k3`	1.0M	$0.95/M	$14.00/M
Wafer`wafer/Kimi-K3`	1.0M	$1.44/M	$9.00/M
MakoraSave 15%`makora/moonshotai/Kimi-K3`	1.0M	$2.55/M	$12.75/M
NovitaSave 10%`novita/moonshotai/kimi-k3`	1.0M	$2.70/M	$13.50/M
TogetherAIBYOK only`together_ai/moonshotai/Kimi-K3`	1.0M	$2.70/M	$13.50/M
DeepInfra`deepinfra/moonshotai/Kimi-K3`	1.0M	$2.85/M	$14.25/M
Baseten`baseten/moonshotai/Kimi-K3`	1.0M	$3.00/M	$15.00/M
EmpirioLabs`empiriolabs/kimi-k3`	1M	$3.00/M	$15.00/M
FireworksBYOK only`fireworks_ai/kimi-k3`	1.0M	$3.00/M	$15.00/M
Inference.net`inference_net/moonshotai/kimi-k3`	1.0M	$3.00/M	$15.00/M
Moonshot`moonshot/kimi-k3`	1.0M	$3.00/M	$15.00/M
Nebius`nebius/moonshotai/Kimi-K3`	1.0M	$3.00/M	$15.00/M
Parasail`parasail/moonshotai/Kimi-K3`	1.0M	$3.00/M	$15.00/M
Parasail`parasail/parasail-kimi-k3`	1.0M	$3.00/M	$15.00/M
PerplexityBYOK only`perplexity/kimi-k3`	1.0M	$3.00/M	$15.00/M
TeraBYOK only`tera/moonshotai/Kimi-K3`	1.0M	$3.00/M	$15.00/M

Certified where it counts.

Built to meet the security and privacy standards that enterprise and healthcare teams require.

Pricing

Averaged across 17 providers, in USD per 1M tokens after gateway discounts.

Average input: $2.58/M
Average output: $14.53/M
Average cache read: $0.43/M
Average cache write: $0.39/M

Cost by volume

Total cost from 0 to 1T tokens on the cheapest and most expensive provider.

Highest$3.00/MBaseten, EmpirioLabs, Fireworks, Inference.net, Moonshot, Nebius, Parasail, Perplexity, and TeraLowest$0.72/MRelace

Throughput and latency

30-day averages per provider from requests through the Respan gateway, best first. 8 of 17 providers had traffic in that window.

Output speed

Higher is better

74tokens/sbest, on Baseten

Baseten74 tok/s
Moonshot49 tok/s
Makora36 tok/s

Time to first token

Lower is better

1.2sbest, on Baseten

Baseten1.2 s
Moonshot1.6 s

Response latency

Lower is better

1.6sbest, on Inference.net

Inference.net1.6 s
Relace2.8 s
OpenRouter3.7 s

Benchmarks

85 published scores for Kimi K3 from 15 sources. Each score links to where it was published. Percentages are out of 100. Arena ratings are Elo, marked between the lowest and highest rated model on each leaderboard.