Kimi K3 — Specifications
| Developer | Moonshot AI |
|---|---|
| Type | LLM (MoE, reasoning, multimodal) |
| Modality | Text, Vision → Text |
| Parameters | 2.8T total / 16 of 896 experts active (MoE) |
| Context window | 1M |
| Max output | — |
| License | Open weights (due 2026-07-27) |
| Open weights | Yes |
| Released | 2026-07 |
| Input price | $3.00 /1M |
| Output price | $15.00 /1M |
| API providers | Moonshot (Kimi), OpenRouter |
Run it locally
| VRAM (4-bit) | ~1.4 TB |
|---|---|
| Minimum GPU | Multi-node cluster (e.g. 2× 8-GPU H200 141GB nodes) |
What is Kimi K3?
Kimi K3 is Moonshot AI’s 2.8-trillion-parameter open frontier model — a mixture-of-experts
that activates just 16 of 896 experts per token, with a 1M-token context window, native
vision and always-on reasoning. It scores 57 on the Artificial Analysis Intelligence Index,
edging past Claude Opus 4.8 at 56 and trailing only GPT-5.6 Sol (59) and Claude Fable 5 (60)
— the strongest open-weight result recorded to date. Weights were due 27 July 2026.
The significance is the gap, or rather the absence of one. An open-weight model sitting
within three points of the best closed model on the market removes the last comfortable
argument for paying frontier prices purely on capability grounds. At $3 in / $15 out per
million tokens K3 is also half the blended cost of GPT-5.5 while scoring higher. The catch is
operational: roughly 1.4 TB of VRAM at 4-bit means a multi-node cluster — two 8-GPU H200
141GB nodes or equivalent — so “open” here means auditable and portable across providers, not
runnable in your server room. For nearly everyone, K3 is an API decision that happens to come
with an exit route.
Kimi K3 pricing: API cost per 1M tokens
| Input (per 1M tokens) | $3.00 |
|---|---|
| Output (per 1M tokens) | $15.00 |
| Output/input ratio | 5× |
| Blended (4:1 in:out) | $5.40 per 1M tokens |
What Kimi K3 costs per month
Real monthly spend at a 4:1 input-to-output mix — the ratio a typical chat or RAG workload actually produces.
| Workload | Tokens / month | Cost / month |
|---|---|---|
| Side project | 1M in / 0.25M out | $6.75 |
| Small team | 20M in / 5M out | $135 |
| Production | 200M in / 50M out | $1,350 |
Run your own numbers in the AI API cost calculator.
Cheaper alternatives to Kimi K3
| Model | Blended $/1M | You save |
|---|---|---|
| GLM 5.2 open | $2.00 | 63% cheaper |
| Gemini 3.5 Flash | $3.00 | 44% cheaper |
| Gemini 3.1 Pro | $4.00 | 26% cheaper |
Self-host or pay the API?
Kimi K3 is open-weight, so you can run it yourself. It needs ~1.4 TB of VRAM at 4-bit (Multi-node cluster (e.g. 2× 8-GPU H200 141GB nodes)). Self-hosting only beats the API once your volume is high enough to keep that hardware busy — the self-hosting vs API calculator works out the break-even point for your token volume.
Frequently asked questions
How much does Kimi K3 cost per 1M tokens?
Kimi K3 costs $3.00 per 1M input tokens and $15.00 per 1M output tokens. At a typical 4:1 input-to-output mix that blends to about $5.40 per 1M tokens.
How much does Kimi K3 cost per month?
A small-team workload of 20M input and 5M output tokens a month costs about $135 on Kimi K3. A side project (1M in / 0.25M out) costs roughly $6.75.
What is a cheaper alternative to Kimi K3?
GLM 5.2 is the strongest cheaper option in our database at $2.00 per 1M blended — about 63% less than Kimi K3. It is also open-weight, so self-hosting is an option.
Can I run Kimi K3 locally?
Yes. Kimi K3 is open-weight and needs about ~1.4 TB of VRAM at 4-bit quantisation (Multi-node cluster (e.g. 2× 8-GPU H200 141GB nodes)).
Why does Kimi K3 charge more for output than input?
Output tokens are generated one at a time and cannot be batched the way a prompt can, so they cost the provider more to serve. Kimi K3 charges 5× more for output, which is why prompt-heavy workloads are far cheaper to run than generation-heavy ones.
Prices are the published list rates for the model's primary API and are reviewed as providers change them. Volume, batch and cached-input discounts are not included. Compare every model side by side in the AI models database or the LLM leaderboard.
