Moonshot AI on October 23, 2026 released Kimi K3.1 Pro, which the Beijing-based company calls the first model in its new K3.1 series and its most capable 3T-class open-weight model to date. The model and a faster sibling, Kimi K3.1 Pro Highspeed, are available now on Kimi.ai and the Kimi mobile apps, the Kimi Work desktop app, Kimi Code, and the Kimi API. Moonshot says the full model weights will be released by November 4, 2026, so for now the model can only be used through Moonshot's own services.

Moonshot describes K3.1 Pro as a post-trained version of Kimi K3 rather than a new base model, trained with substantially more reinforcement-learning compute across hundreds of downstream domains. According to Moonshot and Artificial Analysis, it keeps K3's mixture-of-experts design of 2.8 trillion total and 104 billion active parameters and its 1-million-token context window, and accepts image as well as text input. Moonshot says the Highspeed variant runs two to three times faster than the standard model.

On Moonshot's own figures, K3.1 Pro scores 65.7% on Terminal-Bench 4.0, a test of coding agents working in constrained terminal environments, ahead of GPT-6 Astra at 58.2% and just behind Claude Opus 5.5 at 66.4%. The company also reports 92.7% with tools on the BenchCAD 3D-reconstruction test, against 96.2% for Claude Opus 5.5 and 95.9% for GPT-6 Astra, and a state-of-the-art 64.4 on OfficeQA Pro v2, a financial-document analysis benchmark. Several of Moonshot's other results come from internal benchmarks and harnesses the company chose, and its footnotes note that some competitor scores were taken from other companies' reports rather than run side by side.

Independent measurement is more limited but broadly supportive. Artificial Analysis, which benchmarked the model at launch, places K3.1 Pro fourth on its Intelligence Index with a score of 55, up 11 points from Kimi K3 and behind only Claude Opus 5.5 at 58 and Claude Sonnet 5.5 and Muse Wonder Full at 56. The firm says that once the weights are published, K3.1 Pro would be the highest-scoring open-weights model and would narrow the gap between leading open and proprietary models on its index from 12 points to 3. Artificial Analysis also measured large gains in agentic work, including 1,846 Elo on GDPval-AA, third overall, while gains on knowledge and science tests such as Humanity's Last Exam were more modest.

The most unusual part of the release is a disclosed trade-off on cybersecurity. Moonshot says it found cyber-related material embedded in its general training data and reinforcement-learning tasks, and removed it from both stages for the open-weight release while keeping general coding work. On its internal xdev-bench-hard exploitation test, a checkpoint trained without those measures solved 90.5% of tasks, while the released model reached 81.0%. Moonshot says it does not intend to release open weights that extend the absolute frontier of cyber capability, and that K3.1 Pro still trails frontier proprietary models on offensive cyber tasks. Reuters, reporting on the curbs, notes that the results have not been independently verified.

Moonshot also reports a lower claim-level hallucination rate on its internal Kimi Hallu Bench Hard, 8.3% versus 23.4% for K3, while cautioning that the test is a deliberately difficult stress test and does not represent everyday error rates. Artificial Analysis's own hallucination measure moved less dramatically, from 53% to 44%.

On pricing, Artificial Analysis lists the standard API at $2 per million input tokens, $10 per million output tokens, and $0.20 per million cached input tokens, which it says is 33% lower than Kimi K3 on all three rates. Moonshot's API documentation says access to the flagship model unlocks after a minimum $1 account top-up, and Kimi Enterprise administrators must switch the model on for their organizations because access is off by default at launch. Moonshot says Zero Data Retention is available for API customers.

DigiEditorial treats Moonshot's October 23 technical blog as the primary source and corroborated the release date, availability, parameter counts, weights timeline, and cyber-capability figures against Artificial Analysis's launch benchmarks and Reuters' reporting. Benchmark claims attributed to Moonshot are the company's own, and the open-weight status remains a promise until the files are actually published.