- Moonshot AI introduced Kimi K3.1 Pro this week, the first model in a new K3.1 series, alongside a faster companion variant called Kimi K3.1 Pro Highspeed.
- The model uses a 2.8-trillion-parameter mixture-of-experts architecture with 104 billion active parameters and a 1-million-token context window.
- On the Terminal-Bench 4.0 coding-agent benchmark, K3.1 Pro scored 65.7%, ahead of GPT-6 Astra’s 58.2% and narrowly behind Claude Opus 5.5‘s 66.4%.
- API pricing is $2 per million input tokens and $10 per million output tokens, down from K3’s $3/$15, with full open weights due by Nov. 4, 2026.
What Happened
Moonshot AI released Kimi K3.1 Pro, a 2.8-trillion-parameter mixture-of-experts model with 104 billion active parameters and a 1-million-token context window, as the first entry in its new K3.1 series. MarkTechPost covered the release on Oct. 23, 2026, alongside a faster companion variant, Kimi K3.1 Pro Highspeed.
Why It Matters
The release intensifies competition among frontier models at a time when OpenAI is pushing GPT-6 Sol and Luna on a lower-error, lower-cost basis and DeepSeek continues raising capital despite a US federal ban. Moonshot AI is positioning K3.1 Pro as evidence that open-weight labs can match closed frontier models on coding-agent tasks.
Technical Details
On Terminal-Bench 4.0, which tests coding agents on realistic, resource-constrained terminal tasks, K3.1 Pro scored 65.7%, ahead of GPT-6 Astra at 58.2% and just behind Claude Opus 5.5 at 66.4%. API pricing is $2 per million input tokens and $10 per million output tokens, down from K3’s $3 and $15, with cached input priced at $0.20 per million tokens.
Who’s Affected
Developers building coding agents and other long-context applications gain a cheaper open-weight option that benchmarks close to Claude Opus 5.5. Moonshot AI’s competitors in open-weight and closed frontier models face renewed pricing and performance pressure.
What’s Next
Moonshot AI says full model weights for Kimi K3.1 Pro will be released by Nov. 4, 2026, letting independent researchers verify the benchmark claims against the closed API version.
