Kimi
Moonshot AIKimi K3.1 Pro and K3 have first-party API lanes; K2.6/K2.7 Code also publish BatchJob rates, with separate Global and Data Zone Foundry rows.
Kimi K3.1 Pro, K3, K2.6, and K2.7 Code are available directly from Moonshot's API. K3.1 Pro has a faster Highspeed tier; K2.6 and K2.7 Code also have official BatchJob prices (60% of standard), and K2.7 Code has a separate Highspeed tier. Moonshot says new K2 API account registrations end October 31, while existing accounts are unaffected by that cutoff; full K2 API retirement is planned by the end of December. This notice is for Moonshot API access, not Microsoft Foundry deployments. Kimi K3 is also available through a Fireworks-hosted Foundry Data Zone listing. These chat models are generation-only, so pair them with an embeddings model for retrieval.
Pricing
per 1M tokens · USD / CHF| Model | Tier / Host | Context | Input | Cached | Output | Confidence |
|---|---|---|---|---|---|---|
Kimi K3.1 Pro Moonshot's new K3.1 series flagship. Cache writes are billed separately at $2/M for a 5-minute TTL or $4/M for a 1-hour TTL; these fees are outside the site's shared rate schema. No K3.1 Foundry rate is listed. | Direct API Kimi direct API | 1.0M | $2.00CHF 1.61 | $0.20CHF 0.161 | $10.00CHF 8.05 | official Moonshot's official Kimi Model Inference Pricing page, captured 2026-10-25, lists `kimi-k3.1-pro` at $2/M cache-miss input, $0.20/M cache-hit input, and $10/M output. It lists cache writes at $2/M for 5-minute TTL and $4/M for 1-hour TTL. The official Kimi K3.1 Pro guide identifies the model and documents its 1,048,576-token context. The Microsoft Foundry Kimi pricing table lists K2-family models only, and an Azure Retail Prices query for Foundry meters containing `k3.1` returned no rows on 2026-10-25. The official docs do not give a release or rate-effective date; 2026-10-25 is the first verified capture date. The Highspeed service tier is recorded as a separate variant below. |
Highspeed$4.00 / $0.20 / $20.00 | ||||||
Kimi K3 Flagship long-horizon coding and knowledge-work model; max reasoning effort at launch. Cache writes cost $3/M at the default 5m TTL or $6/M at 1h. TTL-specific cache-write fees are outside this catalog's shared rate schema. | Direct API Kimi direct API | 1M | $3.00CHF 2.42 | $0.30CHF 0.241 | $15.00CHF 12.08 | official Moonshot's official Kimi API pricing page, rechecked 2026-10-05, lists Kimi K3 at $3/M cache-miss input, $0.30/M cache-hit input, $15/M output, and $3/M cache write. Its context-caching guide lists the optional 1h cache write at $6/M, with 5m as the default TTL, and says the default write split preserves total cost for existing requests. Cache-write/TTL fees are not represented in this catalog's shared rate schema. |
Kimi K2.6 First-party API standard lane. BatchJob is a separate Direct API service tier, not a Foundry deployment. New K2 API account registrations end October 31, 2026; existing accounts are unaffected by that cutoff, and full K2 API retirement is planned by the end of December. | Direct API Kimi direct API | 262.1K | $0.95CHF 0.765 | $0.16CHF 0.129 | $4.00CHF 3.22 | official Moonshot's official Kimi Model Inference Pricing page (platform.kimi.ai/docs/pricing/chat-k26), captured 2026-09-24, lists kimi-k2.6 at $0.95/M cache-miss input, $0.16/M cache-hit input, and $4/M output with a 262,144-token context. The page gives no rate effective date. |
Batch$0.57 / $0.10 / $2.40 | ||||||
Kimi K2.7 Code First-party API standard lane. BatchJob and Highspeed are Direct-only service tiers, not Foundry deployment tiers. New K2 API account registrations end October 31, 2026; existing accounts are unaffected by that cutoff, and full K2 API retirement is planned by the end of December. | Direct API Kimi direct API | 262.1K | $0.95CHF 0.765 | $0.19CHF 0.153 | $4.00CHF 3.22 | official Moonshot's official Kimi Model Inference Pricing page (platform.kimi.ai/docs/pricing/chat-k27-code), captured 2026-09-24, lists kimi-k2.7-code at $0.95/M cache-miss input, $0.19/M cache-hit input, and $4/M output with a 262,144-token context. The page gives no rate effective date. |
Batch$0.57 / $0.114 / $2.40Highspeed$1.90 / $0.38 / $8.00 | ||||||
Kimi K2.6 Native Microsoft Foundry Global tier; cached input now officially metered. Kimi's Direct API is listed separately above. | Foundry ·Global | — | $0.95CHF 0.765 | $0.16CHF 0.129 | $4.00CHF 3.22 | official Input $0.95/M and output $4.00/M official. Azure Retail Prices API 'K2.6 cached glbl' meter now publishes cached input at $0.00016/1K ($0.16/M), effective 2026-07-01, captured 2026-07-23 — replacing the earlier ~$0.19/M billing-reconciled estimate. |
Kimi K2.7 Code Native Microsoft Foundry Global tier. Batch and Highspeed are first-party API service tiers and are listed only on the separate Direct row above. | Foundry ·Global | 262.1K | $0.95CHF 0.765 | $0.19CHF 0.153 | $4.00CHF 3.22 | official Azure Retail Prices API, productName 'Azure Kimi', captured 2026-09-24: 'K2.7 Code Inp glbl Tokens' $0.00095/1K, 'K2.7 Code cached glbl Tokens' $0.00019/1K, and 'K2.7 Code Outp glbl Tokens' $0.004/1K, all effective 2026-07-01. The same rates cover 35 commercial Global regions; two US Government rows are 1.25x. Kimi's official model pricing page independently lists the matching Global-standard rates. |
Kimi K2.5 Thinking Cached input now officially metered. | Foundry ·Global | — | $0.60CHF 0.483 | $0.10CHF 0.081 | $3.00CHF 2.42 | official Input $0.60/M and output $3.00/M official. Azure Retail Prices API 'K2.5 cached glbl' meter publishes cached input at $0.0001/1K ($0.10/M), effective 2026-07-01, captured 2026-07-23. |
Kimi K3 New Fireworks-hosted Microsoft Foundry Global lane. Its rates match Fireworks' own direct serverless rates; the separate Data Zone lane below carries the 10% premium. | Foundry ·Global Fireworks-hosted | 1M | $3.00CHF 2.42 | $0.30CHF 0.241 | $15.00CHF 12.08 | official Azure Retail Prices API, productName 'Azure Fireworks Models', captured 2026-09-17: effective 2026-09-01 meters 'FW Kimi-K3 Gl Inp Tokens' $0.003/1K ($3.00/M), 'FW Kimi-K3 Gl Cd Inp Tokens' $0.0003/1K ($0.30/M), and 'FW Kimi-K3 Gl Opt Tokens' $0.015/1K ($15.00/M), uniform across 20 commercial Global regions. Fireworks' official Kimi K3 model page identifies the Moonshot AI model as ready for serverless inference at the same $3.00/$0.30/$15.00 per M rates with a 1.04M-token context. |
Kimi K3 Fireworks-hosted Microsoft Foundry Data Zone counterpart to the Global lane above. Its 10% premium matches the published Fireworks regional pricing pattern. | Foundry ·Data Zone Fireworks-hosted | 1M | $3.30CHF 2.66 | $0.33CHF 0.266 | $16.50CHF 13.28 | official Azure Retail Prices API, productName 'Azure Fireworks Models', meters 'FW Kimi K3 Inp DZ Tokens' ($0.0033/1K = $3.30/M), 'FW Kimi K3 Cd Inp DZ Tokens' ($0.00033/1K = $0.33/M) and 'FW Kimi K3 Opt DZ Tokens' ($0.0165/1K = $16.50/M), effective 2026-08-01, captured 2026-08-21. One uniform price group across all 20 commercial Data Zone regions, no US-Gov or rounding outlier. Exactly 1.10x the new Fireworks Global lane and Moonshot's direct rate ($3.00/$0.30/$15.00), the same Data Zone premium every other Fireworks-hosted lane on Foundry carries. |
Kimi K2.6 Third-party Fireworks-hosted listing with a published cached column. | Foundry ·Data Zone Fireworks-hosted | — | $1.04CHF 0.841 | $0.176CHF 0.142 | $4.40CHF 3.54 | official Fireworks live pricing page. |
Quirks & gotchas
The cache meter the pricing page omitted — now published
Microsoft Foundry's public page long showed no cached column for the K2.6 Global listing, so the rate was back-solved from a Cost Management export at ~$0.19 / CHF 0.15 per M (about 81% hit rate). As of 2026-07-01 Azure's Retail Prices API publishes a dedicated 'K2.6 cached glbl' meter at $0.16 / CHF 0.13 per M, so the catalog now carries the official figure instead of the reconciled estimate.
Data Zone premium via Fireworks
The Fireworks-hosted K2.6 Data Zone listing ($1.045 / CHF 0.84 input) sits above the native Global tier ($0.95 / CHF 0.77).
K3's other Foundry meters do not name their own dimensions
Alongside the clearly-named Fireworks lane tracked here, Microsoft also published six K3 meters under its native 'Azure Kimi' product on 2026-08-01 — but they are labelled 'Model 6' and 'Model 7' with input/output tags that contradict their own prices. 'Model 7 Outp glbl' bills $0.33/M and 'Model 6 Outp DZ' bills $3.00/M, for instance, and no meter is named as cached input at all.
The six prices do form two coherent sets — $3.00/$0.30/$15.00 across 35 regions and $3.30/$0.33/$16.50 across 19 — which look like a Global and a Data Zone lane, and the second set matches the Fireworks meters exactly. But assigning those numbers to input, cached, and output would mean importing the mapping from Moonshot's own price list and presenting the result as an Azure rate. Until Microsoft renames the meters, only the Fireworks Data Zone lane is quoted here, and no Foundry Global row for K3 is claimed.