Sixty-five point seven, weights November fourth

The stamp on this IT Home page is 2026/10/24 7:08:43. It says Moonshot released Kimi K3.1 Pro in the early hours of that day, the first model in the K3.1 series, and calls it the company's most capable 3-trillion-parameter model so far, with a 1 million token context. K3.1 Pro Highspeed came out at the same time, at the same intelligence, with about two to three times the output speed. Full weights are to be released before November 4. The official blog marks the same post with the date 2026-10-24 and no clock, and says the weights will be released by Nov 4. Before November 4, and by Nov 4, are not weights a person can already download. The official page also says enterprise administrators can turn it on, and that access is off by default at launch.

Several public comparisons match across the two pages. On Terminal-Bench 4.0, K3.1 Pro scores 65.7 percent, GPT-6 Astra 58.2 percent, and Claude Opus 5.5 66.4 percent. The official footnote says this item was run with the Claude-Code harness. The internal Kimi Code Bench 3.0 Long-horizon score is 61.2 percent, against 55.95 percent for K3. BenchCAD with tools is 92.7 percent. The official page says that approaches Opus 5.5 at 96.2 percent and GPT-6 Astra at 95.9 percent. OfficeQA Pro v2 is 64.4, which the official page calls the best current result. On the Kimi Hallu Bench Hard stress test, the claim-level error rate fell from 23.4 percent for K3 to 8.3 percent, and the share of answers with at least one factual error was about 42.4 percent of the predecessor's rate. The page itself says the sample is aimed at cases where models slip, and is not an everyday error rate.

The three finance comparisons do not become one sentence. The official page says financial modeling is 7.2 points above K3, financial research is 11.6 points above, and business operation is about 4.2 times K3. IT Home prints a 14 percent rise for modeling, a 20 percent rise for research, and about 4.2 times K3 for operations. The 4.2 times matches. Fourteen percent is not 7.2 points, and 20 percent is not 11.6 points. This story keeps both units. It does not treat the percentages as a translation of the official points.

The price is the yuan figure printed by IT Home. On the Kimi API, kimi-k3.1-pro is 1.3 yuan per million tokens for a cache hit, 13 yuan for a miss, and 68 yuan for output. The high-speed version is about two to three times as fast on output and twice the price. This story does not convert that into dollars. The later part of the official page discusses a security-research evaluation. This story does not write that method or those scores. The internal walking-simulator example is also left without its training steps. How to switch models inside a product is not written here either.