started · updated
Moonshot AI releases Kimi K3.1 Pro with reduced cyber capabilities
Moonshot AI has launched Kimi K3.1 Pro, a new large language model featuring approximately 2.8 to 3 trillion parameters and a 1-million-token context window. The model is a post-trained version of the previous Kimi K3, utilizing increased computational power and reinforcement learning to enhance performance in coding, 3D modeling, and financial research.
To mitigate risks of autonomous hacking, Moonshot AI deliberately removed cybersecurity-related data from its training corpus and reinforcement learning environments. This decision resulted in the model's vulnerability research score decreasing from 90.5% to 81.0%. The company stated its goal is to avoid releasing open-weight models that push the absolute frontier of cyber-offensive capabilities.
In performance benchmarks, Kimi K3.1 Pro achieved a 65.7% score on Terminal-Bench 4.0, outperforming OpenAI’s GPT-6 Astra (58.2%) and nearing Anthropic’s Claude Opus 5.5 (66.4%). The company also introduced a Highspeed variant that operates two to three times faster than the standard version. Open-weight model files are scheduled for release by November 4, 2026.
Following the launch, Moonshot AI issued an apology to users regarding unexpected rate limits and excessive quota consumption caused by sub-agent calls. The company compensated subscribers by restoring 25% of their monthly usage quotas and implemented adjustments to reduce unnecessary sub-agent calls and output lengths.
Entities
Anthropic · Artificial Analysis · Google Cloud · Kimi K3.1 Pro · Linux Foundation · Moonshot AI · OpenAI · Vitalik Buterin
Claims
What the coverage asserts, and how many sources carry each claim.
- [● 12 SOURCES] Moonshot AI removed cybersecurity-related data from the training corpus to limit offensive capabilities. basic-tutorials.de · www.itmedia.co.jp · basic-tutorials.com · cafebiz.vn · thebridge.jp · +6 more
- [● 5 SOURCES] The API price for Kimi K3.1 Pro is $2 per million input tokens and $10 per million output tokens. basic-tutorials.de · www.itmedia.co.jp · basic-tutorials.com · yesky.com · thebridge.jp
- [● 10 SOURCES] The Kimi K3.1 Pro model has approximately 3 trillion parameters. www.itmedia.co.jp · basic-tutorials.com · yesky.com · thebridge.jp · itreseller.pl · +4 more
- [● 6 SOURCES] The model's vulnerability research score decreased from 90.5% to 81.0% due to data filtering. yesky.com · cafebiz.vn · thebridge.jp · itreseller.pl · webrazzi.com · +1 more
- [● 7 SOURCES] Moonshot AI released the Kimi K3.1 Pro model on October 23, 2026. basic-tutorials.de · www.itmedia.co.jp · basic-tutorials.com · businessmag.al · www.cryptopolitan.com · +1 more
- [● 12 SOURCES] Open-weight model files are scheduled to be released by November 4, 2026. basic-tutorials.de · www.itmedia.co.jp · basic-tutorials.com · yesky.com · cafebiz.vn · +6 more
- [● 9 SOURCES] The Kimi K3.1 Pro Highspeed variant is two to three times faster than the standard version. basic-tutorials.de · basic-tutorials.com · yesky.com · thebridge.jp · itreseller.pl · +3 more
- [● 9 SOURCES] The model supports a context window of 1 million tokens. basic-tutorials.de · www.itmedia.co.jp · basic-tutorials.com · yesky.com · thebridge.jp · +3 more