Moonshot shipped Kimi K3.1 Pro on October 23, 2026, and it's now sitting at #4 on Artificial Analysis's Intelligence Index with a score of 55, three points behind Claude Opus 5.5. The API opened at $2 per million input tokens and $10 per million output, with cached input at $0.20, roughly a third below Kimi K3's $3/$15, and 50% below Opus 5.5 and Muse Wonder at $4/$20. Full weights are scheduled for November 4.

That's the narrowest the open-weight-to-proprietary gap has looked on a credible third-party index. It's also the first time a Chinese lab has priced a near-frontier post-trained model this aggressively against the US stack: 80% below GPT-6 Astra's $10/$50 base, and in VentureBeat's October 24 cross-vendor check, DeepSeek V4.1-Flash-1017 was still roughly 6.7× cheaper on uncached input at peak rates. The frontier isn't collapsing to one price; it's fanning out into a tiered open-weight shelf where Moonshot now shares air with Reflection AI's Beam 1.0 and DeepSeek's identically-priced V4.1-Flash refresh.

The architecture is a 2.8-trillion-parameter MoE with 104 billion active parameters, 16 of 896 experts activated per token, a 1-million-token context, and joint RL training spanning hundreds of downstream domains. Thinking mode is mandatory; the API returns HTTP 400 if you try to turn it off. A Highspeed variant priced at $4/$20 with $0.20 cached input promises two to three times the speed at equivalent intelligence.

The benchmarks tell two overlapping stories. On third-party Artificial Analysis measurements, K3.1 Pro posts 67% on Terminal-Bench 4.0 (versus Sonnet 5.5's 64% and K3's 13%), 73% on AutomationBench-AA against Opus's 70%, and 31% on GDP.pdf against Opus's 26%, with a GDPval-AA Elo of 1,846. Opus still leads on the harder reasoning sets: 61% to 52% on Humanity's Last Exam, 32% to 25% on CritPt. Moonshot's own numbers are tighter and friendlier, 65.7% Terminal-Bench against Opus's 66.4% and Astra's 58.2%, 51.7 on SWE-Marathon versus 46.3 and 42.5, 61.2 on Kimi Code Bench 3.0 Long-horizon up from K3's 55.95. MarkTechPost flags the asymmetry: 38.1 on Terminal-Bench Science 0.1, against Astra's 68.1 and Opus's 63.3.

For a small-business owner evaluating AI-powered sales and marketing work, the token-sheet view is the one that lands. LemonLime, which prepares finished outreach, lead research, and content drafts proactively, prices against the quality floor of near-frontier capability. When that floor gets 33% cheaper and the open-weight shelf expands, the economics of done-for-you customer growth tighten in the service's favor. Compression at the frontier doesn't automatically mean compression at the retail layer, but it changes what a $999/month subscription can honestly deliver, and that's before the Mistral Large 4 Le Chonk preview arrives at $0.68/M.

Weights ship November 4. Until then, the index position is a claim with a receipt date attached.

Sources

https://www.kimi.ai/blog/kimi-k3-1-pro
https://venturebeat.com/technology/moonshots-kimi-k3-1-pro-flirts-with-open-weights-leadership-amid-paid-influencer-scandal-allegations
https://thenextweb.com/news/moonshot-kimi-k3-1-pro-open-weight-cyber
https://www.marktechpost.com/2026/10/23/moonshot-ai-kimi-k3-1-pro-open-weight-model/
https://www.unite.ai/moonshot-ai-releases-kimi-k3-1-pro-first-model-in-new-k3-1-series/