In short
- Launched; we read it as GA. The pricing page has a row for the model and the banner reads "Kimi K3.1 Pro has launched!", with no preview, waitlist or limited-access wording. The page never uses the term GA, so this is our reading; the table below is that row, verbatim.
- The API model name is
kimi-k3.1-pro, and a faster same-quality variant,kimi-k3.1-pro-highspeed, sits in the same table. - Standard pricing is $2.00 input / $10.00 output / $0.20 cached input per 1M tokens. The previous generation Kimi K3 is $3.00/$15.00, so K3.1 Pro is the cheaper model.
- Context is 1,048,576 tokens for the whole K3 series (verbatim: "All K3 series models support a 1,048,576-token context window.").
- Cache billing is on by default (default TTL 5 minutes). Writes are priced separately for the 5-minute and 1-hour TTLs.
- Batch API is 60% of the standard price for
kimi-k3.1-proandkimi-k3.
Pricing (USD per 1M tokens, official pricing page)
| Model | Input | Output | Cached input | 5m writes | 1h writes |
|---|---|---|---|---|---|
| kimi-k3.1-pro New | $2.00 | $10.00 | $0.20 | $2.00 | $4.00 |
| kimi-k3.1-pro-highspeed | $4.00 | $20.00 | $0.20 | $4.00 | $8.00 |
| kimi-k3 (previous generation) | $3.00 | $15.00 | $0.30 | $3.00 | $6.00 |
Official notes: "1 MTok = 1,000,000 tokens. All prices above are per MTok consumed." and "Prices exclude applicable taxes."
Worked example (standard rates, no cache)
- 100,000 input tokens + 20,000 output tokens = $0.40 ($0.20 input + $0.20 output).
- The same workload on Kimi K3 costs $0.60 ($0.30 + $0.30), so K3.1 Pro is two-thirds of the cost.
Source: Kimi API Platform — Model Inference Pricing (verified 2026-11-07)
The Chinese-language page is priced in yuan (CNY)
Kimi publishes two official docs sites, platform.kimi.com (Chinese) and platform.kimi.ai (English), and the currency differs. The Chinese table is denominated in yuan, so do not compare those numbers with US-dollar tables without converting.
| Model | Input | Output | Cached input | 5m writes | 1h writes |
|---|---|---|---|---|---|
| kimi-k3.1-pro | ¥13.00 | ¥68.00 | ¥1.30 | ¥13.00 | ¥26.00 |
| kimi-k3.1-pro-highspeed | ¥26.00 | ¥136.00 | ¥1.30 | ¥26.00 | ¥52.00 |
| kimi-k3 (previous generation) | ¥20.00 | ¥100.00 | ¥2.00 | ¥20.00 | ¥40.00 |
Source: Kimi API 开放平台 — 模型推理价格说明 (verified 2026-11-07). This site's comparison table and calculator use the USD figures from the English page.
How cache billing works (automatic, 5-minute and 1-hour TTLs)
The Kimi API turns on context caching billing automatically for repeated request prefixes. Verbatim: "The Kimi API automatically enables context caching billing for repeated request prefixes, with a default TTL (time-to-live) of 5m." So this is not an opt-in feature that you enable per request; workloads that resend the same prefix get cache billing without any configuration.
- Cached input is $0.20, a tenth of the $2.00 input rate. Cached tokens are billed at the cache-hit price only, with no additional cache write charge (verbatim: "cached input is billed at the cache-hit price only, with no additional cache write charge").
- Writes are priced by TTL: $2.00 for the 5-minute tier and $4.00 for the 1-hour tier. Holding a prefix longer costs more.
- Each hit refreshes the cache entry's TTL automatically.
- See the official context caching best practices for choosing a TTL tier.
Related: What prompt caching is · Token cost calculator
Batch API is 60% of the standard price
Verbatim: "kimi-k3.1-pro and kimi-k3 support batch inference for large-scale tasks that do not require real-time responses, billed at 60% of the standard price." For non-realtime bulk work that implies roughly $1.20 input and $6.00 output, but 60% is our arithmetic; the official BatchJob pricing page is the authoritative table.
Specifications (what is published, what is not)
| Item | Detail | Source |
|---|---|---|
| API model name | kimi-k3.1-pro / kimi-k3.1-pro-highspeed | Official pricing page |
| Context | 1,048,576 tokens (K3 series) | Official pricing page |
| Max output | Not found on the pricing page (other official docs not fully reviewed) | — |
| Positioning | "flagship model for long-horizon coding and end-to-end knowledge work" | Official site description |
| Billing model | Pay-as-you-go; no subscription tier (separate from Kimi membership and Kimi Code) | Official pricing page |
| Capabilities | Long context, multimodal understanding, Tool Calling | Official site description |
| Knowledge cutoff / parameter count | Not verified by this site | — |
Note: Kimi Code's 2026-10-23 announcement described the model as "our most capable 3T-class model to date", but the API pricing page states no parameter count, so the table above only carries what the pricing page supports.
Compared with Kimi K3
- K3.1 Pro is cheaper on every line: $2.00 vs $3.00 input and $10.00 vs $15.00 output, about 33% lower on both.
- Context is identical at 1,048,576 tokens, a K3-series-wide limit.
- Cache rates are lower too: $0.20 vs $0.30 cached input, $2.00 vs $3.00 for 5-minute writes, $4.00 vs $6.00 for 1-hour writes.
- K3.1 Pro adds a Highspeed variant (same quality, faster, 2x the price).
Related: Full pricing comparison · China AI overview · Release timeline
K2 models retire at the end of December 2026
Official note (verbatim): "starting October 31, 2026, K2 series models will no longer be available to newly registered users; existing users are not affected. We plan to retire all K2 series models by the end of December 2026". If you still call kimi-k2.7-code or kimi-k2.6, plan the migration to K3 series now.
Availability
- Launched (we read it as GA): the pricing page lists the model and the banner reads "Kimi K3.1 Pro has launched!", with no preview, waitlist or limited-access wording. The page itself does not use the term GA.
- Distribution: Kimi API Platform (platform.kimi.ai / platform.kimi.com) on pay-as-you-go billing; also offered through Kimi Code.
- Cloud marketplaces (e.g. Amazon Bedrock) are not verified on this page. Kimi K3 is available on Bedrock, so check that separate billing tier if you route through it.
Sources: Model Inference Pricing · Kimi K3.1 Pro model guide (verified 2026-11-07)
Where it fits, and what to watch
- Good for repeated long prefixes. Because caching is automatic, agent workloads that resend the same system prompt or repository context pay below the $2.00 headline rate.
- One-shot huge inputs get little cache benefit. Keeping the prefix stable beats streaming a different 1M-token document each call.
- Send non-realtime batches through Batch API for the 60% rate.
- Watch the currency. The Chinese page is in yuan, the English page in USD; our tables and calculator use USD.
- Plan the K2 migration before the end of December 2026.
Related: Pricing · Calculator · Timeline · Prompt caching · China AI