Kimi K2 Thinking
The original Kimi K2 Thinking model, an open-weight reasoning MoE, available via inference gateways.
Source: Moonshot AI pricing, Kimi models
Pricing
Every billing leg from the live catalog, in USD per 1M tokens.
| Billing leg | Rate | Batch |
|---|---|---|
| Input | $0.6 / M | $0.3 / M |
| Output | $2.5 / M | $1.25 / M |
Cost calculator
Billing-grade math over the rate card above, including the thinking tokens most estimates miss.
Hidden reasoning tokens bill as output. 1.0x = thinking off. Medium-effort reasoning typically lands near 3x; heavy reasoning runs 6x to 8x. Set it to match your workload.
Compare providers
The same model, priced across every platform that serves it. Lowest combined in/out rate is flagged.
| Platform | Input / M | Output / M | Cache read / M | Batch | Access |
|---|---|---|---|---|---|
AWS BedrockLowest | $0.6 | $2.5 | — | ✓ | Direct |
Azure AI Foundry | $0.6 | $2.5 | — | — | Direct |
Google | $0.6 | $2.5 | — | — | via OpenRouter |
Novita | $0.6 | $2.5 | $0.15 | — | via OpenRouter |
AWS Bedrock · Regional | from$0.6 | $2.5 | — | ✓ | Direct |
Azure AI Foundry · Data zone | $0.66 | $2.75 | — | — | Direct |
Regional deployments are priced per region; the lowest available region rate is shown.
Price history
Has this model gotten cheaper or more expensive?
| Effective | Input / Output | Status |
|---|---|---|
| July 6, 2026 – July 8, 2026 | $0.6 / $2.5 | past |
| July 8, 2026 | $0.6 / $2.5 | current |
More from Moonshot AI (Kimi)
Other models from the same provider.
Stop estimating. Track it.
SuperPenguin meters every Kimi K2 Thinking call your team makes, prices it with this exact rate card, and shows spend by feature, customer, and team.
Rates are generated from SuperPenguin's live pricing catalog.