Kimi K3.1: What the Pi Post Signals, and What Is Still Unknown
Kimi's official Zhihu account posted digits of pi that the community reads as a K3.1 signal. Here is what is actually known about Kimi K3.1, what remains unconfirmed, and which Kimi APIs you can use today.
Published 2026-09-20
Kimi has not officially released or documented Kimi K3.1. On September 19, 2026, Kimi's verified Zhihu account posted a long digit string beginning 415926535897932.... It matches digits of pi after 3.1, which is why the community widely reads the post as a K3.1 teaser. It is a meaningful official-channel signal, but it is not an API announcement: Kimi has published no K3.1 model ID, release date, parameter count, context limit, price, benchmark sheet, or open-weight commitment.
The newest documented Kimi Code update is still K2.8 Preview, released on September 11. It is a coding-service update on the kimi-for-coding route, not proof that K3.1 is live or that K2.8 has been renamed. For a Kimi API you can call today, APIMaster has kimi-k3 from $2.25 input / $11.25 output per 1M tokens across six active routes, versus Kimi's published $3 / $15 reference price (about 25% off, checked September 20, 2026). See the live Kimi K3 marketplace card; route prices and availability can change.
Updated September 20, 2026. This is a tracking page, not a K3.1 launch announcement.
GPT Trial(GPT-6 Astra, GPT-5.6, GPT-5.5, and GPT Image 2 available)
Sign up and claim a $20 GPT trial
Is Kimi K3.1 released?
No. There is no public Kimi K3.1 API or official product page as of this update. The pi post is the closest thing to a first-party K3.1 signal, but it does not establish a shippable model.
| Item | Kimi K3.1 status |
|---|---|
| Official-channel signal | Yes: a pi digit string posted by Kimi's verified Zhihu account on September 19 |
| Official K3.1 launch announcement | Not published |
| Public API model ID | Not published |
| API price or billing rules | Not published |
| Parameters, context, modalities, benchmarks | Not published |
| Release date | Not announced |
| Open weights | Not confirmed |
| Kimi model usable today | kimi-k3 on the public API; K2.8 Preview inside Kimi Code |
Treat a route, benchmark table, pricing claim, or "K3.1 is live" post as unverified unless Moonshot publishes the matching model documentation and a callable identifier.
What did Kimi post, and why do people read it as K3.1?
At 02:21 Beijing time on September 19, Kimi's verified Zhihu organization account posted no explanatory text, only a long number beginning:
415926535897932384626433832795...
That sequence is pi after the leading 3.1: 3.1415926535897932384626433832795.... The interpretation is simple rather than technical: posting the decimal tail after 3.1 lets the account imply 3.1 without printing a model name.
The post deserves attention because it came from an official Kimi account, not an unattributed screenshot or a third-party model directory. But it still leaves every developer-relevant question unanswered. A teaser does not tell us whether K3.1 is a patch release, a new API tier, a Kimi Code-only rollout, an open-weight checkpoint, or merely an internal milestone.
How does K3.1 relate to K2.8 Preview and K3?
These are different tracks and should not be collapsed into one release story.
| Model or signal | Where it appears | Current status | What is confirmed |
|---|---|---|---|
| Kimi K3 | Kimi open platform and Kimi Code | Released | Public API model kimi-k3; 1M context option; published token pricing |
| K2.8 Preview | Kimi Code | Released September 11 | kimi-for-coding route; 1M context; three reasoning levels; Kimi positions it as approaching K3 for coding |
| Kimi K3.1 | Official Zhihu signal | Not released | A pi-based teaser only; no public technical specification |
K2.8 Preview is the concrete update developers can evaluate today. Kimi's release notes say it improves coding and agent capability and thinking efficiency compared with K2.7 Code; it keeps the existing kimi-for-coding identifier. That is useful context for what Moonshot is already working on, but it is not evidence that K3.1 inherits K2.8's specifications or access rules.
K3 remains the released flagship baseline. The official Kimi Code documentation lists K3 with a 1M-token context option, visual input, and low, high, and max reasoning settings. Do not copy any of those facts into a claimed K3.1 spec until Moonshot publishes them for K3.1.
What is the community expecting from Kimi K3.1?
Since late July, discussion around a possible K3.1 has clustered around a maintenance-and-efficiency update rather than a bigger parameter-count announcement. The recurring expectations are faster response, fewer unnecessary reasoning tokens in long tasks, more reliable coding and agent work, and another competitive step against frontier coding models.
Those are useful things to watch for, not announced features. None should be treated as a K3.1 promise:
- Speed and latency: no official throughput, time-to-first-token, or latency target has been published.
- Reasoning-token efficiency: Kimi has described better thinking efficiency for K2.8 Preview, but has not extended that claim to K3.1.
- Coding and agent quality: no K3.1 benchmark, evaluation setup, or product documentation exists.
- Open weights: K3 was released with open weights, but Moonshot has not committed to an open K3.1 checkpoint.
This distinction matters for purchasing decisions. A model can be widely anticipated and still have a different name, price, release surface, or capability mix when it arrives.
What should developers do while waiting for K3.1?
Build around a configuration value, not a guessed model name. Keep the model ID in environment configuration, then make the later switch after the provider documents the exact identifier, pricing, limits, and release stage.
For a practical Kimi baseline today, compare the released choices:
| Model ID on APIMaster | Best starting point | Lowest live price per 1M tokens | Reference price | Active routes |
|---|---|---|---|---|
kimi-k3 |
Large repositories, long tool loops, harder reasoning | $2.25 in / $11.25 out | $3 / $15 | 6 |
kimi-k2.7-code |
Lower-cost coding workloads | $0.71 in / $3.00 out | $0.95 / $4.00 | 5 |
kimi-k2.6 |
General Kimi API workloads | $0.75 in / $3.50 out | $0.75 / $3.50 | 1 |
The prices above were read from APIMaster's live marketplace on September 20, 2026. kimi-k3 is about 25% below the published reference price, and kimi-k2.7-code is about 25% below its reference price. The marketplace card, not this static table, is the source of truth when you send production traffic.
If you use Kimi Code rather than the public API, start with the K2.8 Preview guide. K2.8 Preview belongs to the Kimi Code subscription surface; it is not a public pay-as-you-go kimi-k2.8 model ID.
How to verify a real Kimi K3.1 launch
Before updating a production configuration, check for all of the following:
- A Moonshot or Kimi announcement that names the model and release stage.
- Official API documentation that lists the exact accepted model ID.
- A price table covering input, output, cached input, and any long-context rules.
- Published context and output limits, supported modalities, tool calling, and region or account restrictions.
- A low-volume test that returns the documented identifier through the endpoint you actually use.
An account post, a community leak, or an aggregator label can tell you what to monitor. It cannot replace the model card and API documentation required to plan capacity, cost, or a production migration.
Kimi K3.1 FAQ
Is Kimi K3.1 officially announced?
No. Kimi's verified Zhihu account posted a pi digit string that is widely read as a K3.1 teaser, but Moonshot has not published a K3.1 product announcement, API page, price, or model card.
What is the Kimi K3.1 API model ID?
Unknown. Do not send kimi-k3.1 or another guessed string to an API. Wait for Moonshot's documentation and copy the exact supported model ID from the provider you call.
When will Kimi K3.1 be released?
Kimi has not announced a release date. The September 19 pi post provides a reason to watch official channels, not a calendar commitment.
Will Kimi K3.1 be open weight?
K3's open-weight release does not guarantee that K3.1 will follow the same path. Moonshot has not announced an open-weight commitment for K3.1.
Is K2.8 Preview K3.1?
No. K2.8 Preview is Kimi Code's September 11 update on the kimi-for-coding route. K3.1 is an unannounced name inferred from a later official-account teaser.
Which Kimi API can I use now?
Use kimi-k3, kimi-k2.7-code, or kimi-k2.6 from the current APIMaster Kimi marketplace. Check the live card for routes and prices. K3.1 will be added when Moonshot releases a documented API model and an active route is available.
Sources and verification date
- Kimi verified Zhihu organization account, September 19, 2026: pi digit-string post. This is the direct source for the K3.1 teaser interpretation; it includes no specification or launch details.
- Kimi Code release notes, September 11, 2026: K2.8 Preview release, coding and agent positioning, thinking-efficiency claim, and 1M context access.
- Kimi Code model documentation: current K3 and K2.8 model IDs, access rules, context limits, and reasoning controls.
- Kimi K3 quickstart and pricing: released K3 API baseline.
- APIMaster live marketplace data for
kimi-k3,kimi-k2.7-code, andkimi-k2.6, checked September 20, 2026. Prices and route counts can change.
Start with a released Kimi model
K3.1 is worth tracking, but it is not a model you can configure today. Use a released Kimi model as your baseline, then compare again when Moonshot publishes the K3.1 model card.
Create an APIMaster account, create a key in the console, point your client at https://apimaster.ai/v1, and begin with kimi-k3. When a documented K3.1 route becomes available, this page will be updated with the real model ID, pricing, and limits.