Kimi K3.1 Pro
- Provider
- Moonshot AI
- Status
- Current
- Context
- 1,000,000 tok
- Price
- $2 / $10 /MTok
Kimi K3.1 Pro is Moonshot’s flagship model, released on 23 October 2026 in the Kimi apps and on the Kimi API (Moonshot). Moonshot calls it its “most capable 3T-class open-weight model”, but the weights are not out yet: Moonshot says they will be published by 4 November 2026. On the API it costs $2/$10 per million tokens, with a 1M-token context.
Artificial Analysis scored it 55 on the Intelligence Index v4.3.2, at $2.11 per completed index task. Moonshot published no SWE-bench figure, no exact parameter count and no licence for the coming weights.
Quick specs
| Kimi K3.1 Pro | |
|---|---|
| Released | 23 October 2026 |
| Input price | $2.00 / MTok |
| Output price | $10.00 / MTok |
| Context window | 1,000,000 tokens |
| Parameters | ”3T-class” (exact count not published) |
| Max output | Not published |
| Knowledge cutoff | Not published |
| AA Intelligence Index | 55 |
| Cost per AA index task | $2.11 |
| Output speed | 52.2 tokens a second (Artificial Analysis) |
| API model IDs | kimi-k3.1-pro, kimi-k3.1-pro-highspeed |
| Open weights | Promised by 4 November 2026 |
Prices are as listed by Artificial Analysis, based on Kimi’s API; Moonshot’s launch post links to its pricing page without printing the rates. The Highspeed variant runs two to three times faster, according to Moonshot.
What Moonshot says is new
Coding and long tasks. Moonshot reports 61.2% on its internal Kimi Code Bench 3.0 long-horizon test, up from 55.95% for Kimi K3.
Office and finance work. It claims a state-of-the-art 64.4 on OfficeQA Pro v2, and large gains over K3 on its own finance benchmarks, including Financial Research (70.4 against 58.8).
Fewer hallucinations. On its internal Kimi Hallu Bench Hard, a stress test of deliberately hard prompts, Moonshot reports a claim-level hallucination rate of 8.3% with tools, down from K3’s 23.4%. Moonshot says the test does not reflect typical error rates.
Cyber data removed. Moonshot says it removed cyber-related data from both training stages so the open weights do not advance offensive cyber capability. On its internal xdev-bench-hard test, that lowered the score from 90.5% to 81.0%.
Benchmark performance
Independent, from Artificial Analysis:
| Measure | Kimi K3.1 Pro | Claude Opus 5.5 | Claude Sonnet 5.5 | GPT-6 Astra |
|---|---|---|---|---|
| Intelligence Index v4.3.2 | 55 | 58 | 56 | 53 |
| Cost per index task | $2.11 | $5.98 | $7.60 | $3.26 |
| Price (in/out per MTok) | $2 / $10 | $4 / $20 | $2 / $10 | $10 / $50 |
Artificial Analysis ranks it 4th of 215 models on its page. It used 150 million output tokens to run the index, which Artificial Analysis describes as very verbose against a median of 87 million, and its output speed of 52.2 tokens a second is slow for the tier.
Published by Moonshot, which notes that several of its benchmarks are internal and that some evaluations ran on H20 rather than H100 GPUs:
| Benchmark | Kimi K3.1 Pro | Claude Opus 5.5 | GPT-6 Astra |
|---|---|---|---|
| Terminal-Bench 4.0 (Claude Code harness) | 65.7% | 66.4% | 58.2% |
| BenchCAD, with tools | 92.7% | 96.2% | 95.9% |
| BenchCAD, without tools | 72.8% | 73.0% | n/a |
| Kimi Hallu Bench Hard, without tools (lower is better) | 5.4% | 6.8% | 4.2% |
The Opus 5.5 Terminal-Bench figure is the maximum Moonshot found from Anthropic’s blog or the official leaderboard.
How it compares
vs Kimi K3
Kimi K3 (16 July 2026) costs $3/$15 per million tokens and scores 44 on the same Artificial Analysis index. K3.1 Pro is 11 points higher at a third lower token price. K3’s weights are published; K3.1 Pro’s are due by 4 November.
vs Claude Sonnet 5.5
Claude Sonnet 5.5 has the same $2/$10 token price and scores one point higher on the index (56). K3.1 Pro costs less per completed task, $2.11 against $7.60, because Sonnet 5.5 uses more output tokens at max effort.
vs Claude Opus 5.5
Claude Opus 5.5 scores three points higher on the index (58) at twice the token price, and $5.98 per completed task. On Moonshot’s Terminal-Bench 4.0 table the two are within a point.
Known limitations
No SWE-bench figure. Moonshot published no SWE-bench Pro or Verified result.
Weights not yet released. Moonshot calls it an open-weight model, but until the weights and a licence are published it is available only through Moonshot’s apps and API.
Slow and verbose. Artificial Analysis measured 52.2 output tokens a second and 150 million output tokens to run its index, against a median of 87 million.
Specs withheld. Exact parameter count, knowledge cutoff and maximum output length are not published.
Enterprise default. It is off by default for enterprise accounts until an admin enables it.
Frequently asked questions
When was Kimi K3.1 Pro released?
23 October 2026, on Kimi.ai, the Kimi apps, Kimi Work, Kimi Code and the Kimi API.
How much does Kimi K3.1 Pro cost?
$2 per million input tokens and $10 per million output tokens on the Kimi API, as listed by Artificial Analysis, which measured $2.11 per completed Intelligence Index task.
Is Kimi K3.1 Pro open source?
Not yet. Moonshot says the full weights will be published by 4 November 2026 and has not named the licence. Until then it is available only through Moonshot’s apps and API.
Is Kimi K3.1 Pro better than Claude Opus 5.5?
Not on the independent index: Artificial Analysis scores Kimi K3.1 Pro 55 and Claude Opus 5.5 58. Kimi K3.1 Pro costs half as much per token ($2/$10 against $4/$20), and on Moonshot’s own table it is within a point of Opus 5.5 on Terminal-Bench 4.0 (65.7% to 66.4%).
How is Kimi K3.1 Pro different from Kimi K3?
Kimi K3.1 Pro scores 55 on the Artificial Analysis index against Kimi K3’s 44, at $2/$10 per million tokens against K3’s $3/$15. Moonshot also says it removed cyber-related training data and cut hallucinations on its internal stress test from 23.4% to 8.3%.