Moonshot AI Releases Kimi K3.1 Pro, the First Model in Its K3.1 Series
Moonshot AI, the Beijing-based company behind the Kimi assistant, released Kimi K3.1 Pro in the final week of October 2026. Sources give the announcement date as 23 or 24 October. K3.1 Pro is a heavily post-trained successor to Kimi K3, the open-weight model of nearly three trillion parameters that Moonshot shipped in June 2026. The release comes in two variants: Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed, which Moonshot says runs two to three times faster. Both are available now on Kimi.com, the Kimi mobile apps, the Kimi Work desktop app (version 3.1.0 and later), Kimi Code, and the Moonshot API under the identifiers kimi-k3.1-pro and kimi-k3.1-pro-highspeed. Moonshot has promised to publish the full model weights by 4 November 2026. As of 26 October, no K3.1 Pro repository had appeared on Hugging Face.
Kimi K3.1 Pro Benchmarks: Within a Point of Claude Opus 5.5 on Coding
The headline number is Terminal-Bench 4.0, a benchmark for AI coding agents working in a real terminal. Moonshot reports that Kimi K3.1 Pro scores 65.7%, run with the Claude Code harness. Claude Opus 5.5 scores 66.4% and GPT-6 Astra scores 58.2%, with the competitor figures taken from each lab's published results or public leaderboards. On Moonshot's internal long-horizon coding suite, the score rose from 55.95 for K3 to 61.2 for K3.1 Pro. Moonshot also reports 92.7 on BenchCAD with tools, measured on a 1,000-case subset, and 64.4 on OfficeQA Pro v2.
Every one of these figures comes from Moonshot's own runs, and the harnesses differ between models, so they are not controlled comparisons. The independent Artificial Analysis index scores K3.1 Pro at 55, eleven points above K3. Teams should treat the vendor numbers as a reason to run their own evaluation, not as a substitute for one.
Fewer Hallucinations
Moonshot says K3.1 Pro makes far fewer false claims. In the launch post, the share of false claims on its internal hard test set fell from 23.4% for K3 to 8.3% for K3.1 Pro. The API quickstart quotes a different pair of figures, 23.35% to 6.85%, for what appears to be the same measurement. Either way, Moonshot is reporting a substantial drop in hallucination rate, which matters most for research, document analysis and customer-facing use.
Pricing, Context Window and API Behaviour
Kimi K3.1 Pro costs $2 per million input tokens and $10 per million output tokens, with cached input at $0.20 per million. That is about one-third cheaper on every rate than Kimi K3, which was priced at $3 input and $15 output. The Highspeed variant costs $4 per million input tokens and $20 per million output tokens. The context window is 1,048,576 tokens, shared between input and output, with a default maximum completion of 131,072 tokens that can be raised to the full window. The model accepts image and video input for understanding but does not generate images or video.
Developers moving from other APIs should note three behaviours. Thinking is always on and cannot be disabled; reasoning effort can be set to low, high or max, with max as the default. Sampling parameters are fixed at temperature 1.0 and top-p 0.95, and Moonshot recommends omitting them from requests. Zero data retention is available, and enterprise administrators must enable the model for their organisation because it is off by default.
Moonshot Deliberately Weakened K3.1 Pro's Hacking Skills
Before the open-weight release, Moonshot removed cyberattack-related material from the training data. On Moonshot's in-house test of exploiting real software, a version trained without that filtering scored 90.5%, while the shipped K3.1 Pro scored 81%. K3 scored 33.3% on the same test. Moonshot says its stated aim is to avoid pushing the frontier of what freely downloadable models can do in offensive security, because open weights cannot be recalled once released. Even with the cut, K3.1 Pro is far more capable at exploitation than its predecessor, which is why security researchers continue to argue that open-model safeguards lag behind closed ones.
What Kimi K3.1 Pro Means for Indian Businesses and Developers
For Indian software teams, K3.1 Pro is a near-frontier coding model at a mid-tier price. At $2 input and $10 output per million tokens, it is a practical option for high-volume agentic workloads such as code migration, test generation and document processing, where frontier closed models can become expensive at scale. Once the weights are published, organisations with strict data-residency needs could self-host it inside Indian data centres, keeping prompts and documents under their own control as the Digital Personal Data Protection rules take effect.
The trade-offs are real. Self-hosting a model in the three-trillion-parameter class needs serious GPU infrastructure, the licence terms for K3.1 Pro have not been published, and enterprises in regulated sectors will want to review the US government's July 2026 allegations against Moonshot before adopting it. A sensible path is to benchmark K3.1 Pro against your current model on your own tasks through the API, then decide on self-hosting once the weights and licence are public.
The Bottom Line
Moonshot AI's Kimi K3.1 Pro, released on 23–24 October 2026, is a post-trained upgrade to Kimi K3 with a 1M-token context window, image and video input, and API pricing of $2 per million input tokens and $10 per million output tokens. Moonshot reports 65.7% on Terminal-Bench 4.0, close to Claude Opus 5.5's 66.4%, and a fall in false claims from 23.4% to 8.3%, though all benchmarks are vendor-reported. Open weights are promised by 4 November 2026, and Moonshot deliberately reduced the model's cyber-exploitation skill from 90.5% to 81% before release. For Indian teams, it is a strong, low-cost candidate for coding and document workloads that deserves a proper in-house evaluation.
Frequently Asked Questions
What is Kimi K3.1 Pro?+
Kimi K3.1 Pro is a large language model from Moonshot AI, released on 23–24 October 2026 as the first model in its K3.1 series. It is a heavily post-trained successor to Kimi K3, an open-weight model of nearly three trillion parameters. K3.1 Pro has a 1,048,576-token context window, accepts image and video input, and always uses reasoning (thinking) mode. It is available on Kimi.com, the Kimi apps, Kimi Work, Kimi Code and the Moonshot API, alongside a faster Kimi K3.1 Pro Highspeed variant that Moonshot says runs two to three times faster.
How much does the Kimi K3.1 Pro API cost?+
Kimi K3.1 Pro costs $2 per million input tokens, $10 per million output tokens and $0.20 per million cached input tokens. That is roughly one-third cheaper than Kimi K3, which cost $3 per million input tokens and $15 per million output tokens. The faster Kimi K3.1 Pro Highspeed variant costs $4 per million input tokens and $20 per million output tokens. API access requires a minimum top-up of $1, and the cumulative top-up amount sets the account's rate-limit tier.
How does Kimi K3.1 Pro compare with Claude Opus 5.5 and GPT-6 Astra?+
On Terminal-Bench 4.0, a benchmark for AI coding agents, Moonshot reports that Kimi K3.1 Pro scores 65.7%, compared with 66.4% for Claude Opus 5.5 and 58.2% for GPT-6 Astra. Moonshot ran its own model with the Claude Code harness and took the competitor scores from each lab's published results, so the comparison is not fully controlled. The independent Artificial Analysis index scores K3.1 Pro at 55, eleven points above Kimi K3. Businesses should test the model on their own tasks before relying on these vendor-reported numbers.
Will Kimi K3.1 Pro be open-weight?+
Yes. Moonshot AI has promised to publish the full Kimi K3.1 Pro weights by 4 November 2026. As of 26 October 2026, the weights had not yet appeared on Hugging Face, and the licence terms, file size and hardware requirements had not been announced. Before the open release, Moonshot removed cyberattack-related material from the training data, which reduced the model's score on its internal software-exploitation test from 90.5% to 81%.
Written by
TechPillow Team
Sharing insights on technology, product development, and the Indian tech ecosystem.
