Kimi K3 API Prices Slashed: Input Tokens Drop 71% to $0.884/1M

Moonshot AI has repriced its Kimi K3 API, cutting input costs by 71% — from $3.00 to $0.884 per 1M tokens. Output pricing also fell, moving from $15.00 to $10.53 per 1M tokens, though the reduction there is more modest.

Who Actually Benefits

The input cut is the headline number, and it matters most for workloads that are read-heavy or context-heavy — think RAG pipelines, long-document summarisation, or multi-turn chat where large system prompts are re-sent each turn. At $0.884/1M input tokens, costs on those patterns drop dramatically.

Output-heavy workloads (code generation, long-form drafting) see a real but smaller benefit. Going from $15.00 to $10.53/1M output is meaningful over volume, but developers relying on Kimi K3 primarily for generation will want to run the numbers against alternatives before assuming this is now the cheapest option on the board.

A Pricing Reality Check

The asymmetry between the input cut (-71%) and the output cut (roughly -30%) is worth noting. It suggests Moonshot AI is competing hard for use cases where ingestion cost is the bottleneck — long-context retrieval and analysis workloads — rather than pure generation throughput.

If your token split is roughly 50/50 input/output, blended cost improvement will land somewhere between those two percentages. If you're running 80%+ input-heavy workloads, you're closer to that 71% headline saving.

Bottom Line

This is a genuine cost reduction, not a tier restructure or a limited-time promotion based on available facts. Developers already evaluating frontier long-context models should reprice their estimates accordingly. Check current specs, tier details, and how K3 stacks up against comparable models on Kimi K3 — live specs & price history.

Ezra, Scout AI Team