What is Kimi K3.1 Pro?
Kimi K3.1 Pro is Moonshot AI's flagship model, released on 23 October 2026 in the Kimi apps, Kimi Work, Kimi Code and the Kimi API. It is not a new base model: Moonshot describes it as Kimi K3 with substantially more post-training, using reinforcement learning across hundreds of domains, with gains claimed in coding, finance, agent tasks and factual accuracy. It keeps K3's 1M-token context and image input.
On the API it comes as kimi-k3.1-pro and a Highspeed variant that Moonshot says runs two to three times faster at double the price. Moonshot's launch post says zero data retention is available on the API. Weights are promised by 4 November 2026; until then it is API-only, and the licence has not been published. In organisations it is off by default until an administrator turns it on.
How much does Kimi K3.1 Pro cost?
| Plan | Price | What you get |
|---|---|---|
| K3.1 Pro | $2 in / $10 out per 1M tokens | Cached input $0.20 |
| K3.1 Pro Highspeed | $4 in / $20 out per 1M tokens | Two to three times faster, per Moonshot |
Kimi API: $2 input / $10 output per 1M tokens, cached input $0.20; Highspeed $4 / $20. A third below Kimi K3. Open weights promised by 4 November 2026.
Is Kimi K3.1 Pro worth it?
Kimi K3.1 Pro is the upgrade Kimi K3 needed. Independent testing shows a large jump: Artificial Analysis moved it from 44 to 55 on its Intelligence Index, fourth of 215 models, with Terminal-Bench 4.0 rising from 13% to 67% and its hallucination rate falling from 53% to 44%. Claude Opus 5.5 still leads on the hardest reasoning tests, 61% to 52% on Humanity's Last Exam.
The price fell as the scores rose. At $2 per million input tokens and $10 output it matches Claude Sonnet 5.5 and costs half of Opus 5.5. Artificial Analysis puts it at $2.11 per index task, about what K3 cost for a much weaker result. It is still slow: 57.5 tokens a second on Kimi's API, 141st of 186, which is what the Highspeed variant at $4/$20 is for.
Moonshot's own tables are more flattering and should be read as such. It reports 51.7 on SWE-Marathon against 46.3 for Opus 5.5 and 64.4 on OfficeQA Pro v2, but it trails clearly on Terminal-Bench Science (38.1 against 63.3) and BenchCAD. Several of its finance and hallucination tests are internal.
Two things to weigh before switching. The weights are not out yet; Moonshot promises them by 4 November, and until then this is a closed API model with no published licence. And the launch came with unverified screenshots, reported by VentureBeat, of a creator campaign asking influencers not to label paid posts; Moonshot had not responded as of 25 October. Neither changes what the model does, but both are reasons to rely on independent numbers rather than social media.
Where it earns its place
A team already running agents on Kimi K3
Cut cost and improve results without changing integration
Switch the model ID to kimi-k3.1-pro on the same API. A task with 500K input and 30K output tokens drops from about $1.95 to $1.30 before caching, and Artificial Analysis's agent and terminal scores rise sharply.
A developer comparing Sonnet-priced models for coding agents
Pick a model for long terminal and repository tasks
K3.1 Pro matches Claude Sonnet 5.5 on price and scores 67% on Terminal-Bench 4.0 in Artificial Analysis's run. Use the Highspeed variant where latency matters, at twice the token price.
A company that wants a frontier-class open model on its own hardware
Plan a self-hosted deployment
Wait for the weights and licence, promised by 4 November. Until then, Kimi K3 is the open-weight option, and its cluster requirements are a guide to what K3.1 Pro will need.
Before you commit
- Setup
- Select kimi-k3.1-pro or kimi-k3.1-pro-highspeed on the Kimi API Platform, or use /model in Kimi Code. In organisations, an administrator must enable it first.
- Integrations
- Kimi apps (web, iOS, Android, HarmonyOS), Kimi Work 3.1.0 and later, Kimi Code and the Kimi API. Artificial Analysis lists two API providers at launch.
- Data and privacy
- Moonshot's launch post says zero data retention is available on the API; check the terms before sending sensitive data, since Kimi's general privacy policy allows training on user content. Weights and licence are not yet published.
- Cost at scale
- $2/$10 per million tokens, $0.20 cached, with no long-context surcharge. 1,000 agent tasks of 500K input and 30K output tokens cost about $1,300 uncached, against about $1,950 on Kimi K3.
- Limits
- 1M-token context. Artificial Analysis measured 57.5 tokens a second and a 3.07-second first token on Kimi's API. No self-hosting until the weights ship.
What's good
- Fourth of 215 on Artificial Analysis's Intelligence Index, up from 44 to 55
- $2/$10 per million tokens, a third below Kimi K3
- Hallucination rate down from 53% to 44% in independent testing
- Zero data retention available on the API
What's not
- Weights and licence not released yet
- Slow at 57.5 tokens a second on Kimi's API
- Trails Opus 5.5 on the hardest reasoning tests
- Off by default for organisations until an admin enables it
Best for
Teams running long coding and agent tasks who want near-frontier results at Sonnet prices, and can wait for the weights if they need to self-host.
Kimi K3.1 Pro alternatives
Kimi K3.1 Pro: common questions
- Is Kimi K3.1 Pro worth it?
- The same model as Kimi K3 with far better training on top: eleven points higher on Artificial Analysis's index at a third less per token. Wait for the weights before planning to self-host. We rate it 4.4 out of 5.
- How much does Kimi K3.1 Pro cost?
- Kimi API: $2 input / $10 output per 1M tokens, cached input $0.20; Highspeed $4 / $20. A third below Kimi K3. Open weights promised by 4 November 2026.
- What are the downsides of Kimi K3.1 Pro?
- Weights and licence not released yet; slow at 57.5 tokens a second on Kimi's API; trails Opus 5.5 on the hardest reasoning tests; off by default for organisations until an admin enables it.
- Who is Kimi K3.1 Pro best for?
- Teams running long coding and agent tasks who want near-frontier results at Sonnet prices, and can wait for the weights if they need to self-host.
- How much does Kimi K3.1 Pro cost for a team or in production?
- $2/$10 per million tokens, $0.20 cached, with no long-context surcharge. 1,000 agent tasks of 500K input and 30K output tokens cost about $1,300 uncached, against about $1,950 on Kimi K3.
- Is Kimi K3.1 Pro safe for company data?
- Moonshot's launch post says zero data retention is available on the API; check the terms before sending sensitive data, since Kimi's general privacy policy allows training on user content. Weights and licence are not yet published.
- What are the best Kimi K3.1 Pro alternatives?
- The closest alternatives we have reviewed are Kimi K3, Claude Sonnet 5.5, Claude Opus 5.5, GLM-5.3 and DeepSeek-V4-Pro.
Answers checked Oct 25, 2026.
Make Kimi K3.1 Pro? Show this rating on your site
