Kimi K3 API Pricing (July 2026)

Kimi K3 is Moonshot AI's 2.8-trillion-parameter flagship, released July 16, 2026 — a 1M-token context window, native vision, and pricing that puts it head-to-head with US flagship models. Here is what it actually costs, and how to pay less.

Official Kimi K3 API price

Rates are flat across the full 1,048,576-token context — no long-context surcharge. Prices as of July 2026; check the vendor's pricing page for the authoritative rate.

How to use K3 at 42% below the official rate

gloritoken serves the official, full-quality Kimi K3 API (not a quantized mirror), alongside Qwen 3.8, DeepSeek V4 and GLM 5.2 — one key, up to 48% below the vendors' international rates — K3 itself freshly cut to 42% below (cached input discounted the same), Qwen 3.8 Max at 48% below. It speaks both the OpenAI and Anthropic protocols, so Claude Code connects with two environment variables:

export ANTHROPIC_BASE_URL="https://api.gloritoken.ai"
export ANTHROPIC_AUTH_TOKEN="sk-your-key"
export ANTHROPIC_MODEL="kimi-k3"

Good to know

Demand is high. K3 launched days ago and the upstream engine can return 429 at peak. With gloritoken the same key also runs deepseek-v4-pro or qwen3.8-max, so you can fall back instantly without changing tools.

Is K3 cheap? After our August price cut it undercuts the US flagships on output too — $8.65/1M on gloritoken against Claude Sonnet 5 at $10. Its strengths are the 1M context, vision, and agentic performance. If you want maximum savings, deepseek-v4-flash ($0.18/1M output) or qwen3.7-plus ($0.74) deliver far lower unit costs — full price list here.

gloritoken is an independent platform serving China's frontier models first-hand from the makers' official APIs via our Singapore gateway. Not affiliated with Moonshot AI. We do not relay Claude, GPT or Gemini.