Kimi K3 (Max)Ready
kimi-k3
Serverless APILLMReasoningVisionPrompt cachingOpen weights
A flagship K3 model known for long-context retrieval and tool use. It ranks #14 of the top 20 frontier models on the Artificial Analysis Intelligence Index with a score of 43.6. Released July 16, 2026.
1.0M
Context
2.8T
Parameters
#14
Intelligence rank · 43.6
Open
Weights · Kimi K3 License
View full spec
ReleasedJul 16, 2026
Knowledge cutoff—
Size classlarge
Input modalitiestext, image
Output modalitiestext
Median output speed42 tok/s
Median time to first token3.7s
LicenseKimi K3 License
Usage guide
1. Choose how you'd like to use Kimi K3 (Max)
☁️ Serverless
Use with pay-per-token pricing. Zero setup, no cold starts, automatic failover between upstream routes.
Usage tiers · Standard • Priority
Try it — sign in
curl https://tokenhub-staging.coality.io/v1/chat/completions \
-H "Authorization: Bearer $TOKEN_HUB_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "kimi-k3",
"messages": [{"role": "user", "content": "Explain vector databases in one paragraph"}]
}'Pricing
| Bucket | Price | Notes |
|---|---|---|
| Input tokens | $3.75 | per 1M tokens |
| Output tokens | $18.75 | per 1M tokens |
| Cache read | $0.3750 | cached input · 90% off |
| Cache write | — | per 1M tokens |
Billed per token with micro-dollar precision · effective Oct 26, 2026 · default key limits 600 req/min · 200K tokens/min
Benchmarks
GPQA Diamond
93.5
Humanity's Last Exam
46.9
MMMU-Pro
80.5
SciCode
59.5
Terminal-Bench 4.0
12.6
Long Context Reasoning
88.7
Frontier Code
43.6
Scores from the Artificial Analysis evaluation suite (0–100 scale).
About this model
2.8T-parameter open-weight model, with a 1M context window, native image input, visible reasoning traces, prompt caching at a 90% discount on cached input, 42 output tokens/s median serving speed and 3.7s to first token.