Kimi K3.1 Pro Is Out! Is it a Frontier Model?

Moonshot AI just released Kimi K3.1 Pro. It isn't a new, bigger model: it's the same Kimi K3 from July, with a lot more training on top, and that training took it from 44 to 55 on Artificial Analysis's independent ranking, fourth of 215 models. You'll get what Moonshot shipped, what its own benchmark chart says and leaves out, the independent scores, one prompt on Kimi K3 and K3.1 Pro side by side, what a finished task really costs, and a short comparison with Opus 5.5, GPT-6.1 Sol and the Chinese open models. 🆕 WHAT MOONSHOT SHIPPED • A post-trained version of Kimi K3: same 2.8 trillion parameters, much more training • 1 million tokens of context, text and image input • Live in the Kimi app, Kimi Work, Kimi Code and the API (kimi-k3.1-pro) • $2 per million tokens in, $10 out • Full weights promised by November 4, 2026; until then it runs on Moonshot's servers • Cybersecurity data taken out of its training on purpose: on Moonshot's own hacking test it solves 81% of the problems; the same training without that cut gets 90.5% 📈 THE BENCHMARKS Moonshot's own chart (14 tests): Kimi K3.1 Pro comes first on 6. OfficeQA Pro v2: 64.4, the top score on the chart. Terminal-Bench (the coding-agent test): 65.7%, just behind Claude Opus 5.5's 66.4%. Terminal-Bench Science: 38.1, well behind GPT-6 Astra's 68.1. GPT-6.1 Sol isn't on the chart. Artificial Analysis (independent): Intelligence Index 55, fourth of 215 models, up 11 points from Kimi K3's 44. In its own Terminal-Bench run, Kimi K3 scored 13% and Kimi K3.1 Pro 67%, which Artificial Analysis calls the highest score of any model. One prompt (Peter Gostev, Arena): a 3D dinosaur scene on Kimi K3, K3.1 Pro and GPT-6.1 Sol. A huge jump from K3 to K3.1 Pro; on this prompt, his call is GPT-6.1 Sol. 💭 MY TAKE Kimi K3.1 Pro got this good by training the same Kimi K3 better, not by building a bigger model. Once the weights are out, it's the first open model I'd put next to Opus 5.5 and GPT-6.1 Sol. But Kimi thinks a lot: it costs the same per token as GPT-6.1 Sol, and a finished task costs about three times as much. • Agent and coding work, if you want results close to Opus from an open model: Kimi K3.1 Pro • Cheap everyday work: GPT-6.1 Sol • Only the best result will do: Claude Opus 5.5 • Lowest cost, if slightly lower quality is fine: MiMo-V2.6-Pro or DeepSeek V4.1 Flash Will I use Kimi K3.1 Pro a lot? No. Opus 5.5 is still the best model, and for most things I want the best result. For everyday business work, like answering emails and updating websites, I use GPT-6.1 Sol for the cost. For most people I don't see a use for open-weight models right now: the frontier models give the best results for the money, and the Chinese models are still 3 to 6 months behind the frontier labs. The exception is a business that doesn't want to send its data out. 📊 HOW IT COMPARES Model (maker): Artificial Analysis Intelligence Index · cost per finished index task Claude Opus 5.5 (Anthropic): 58 · $5.98 Kimi K3.1 Pro (Moonshot AI): 55 · $2.11 Claude Fable 5.1 (Anthropic): 53 · $7.63 GPT-6 Astra (OpenAI): 53 · $3.26 GPT-6.1 Sol (OpenAI): 52 · $0.72 MiMo-V2.6-Pro (Xiaomi, open weights): 46 · $0.13 DeepSeek V4.1 Flash 1017 (DeepSeek, open weights): 45 · $0.26 GLM-5.3 (Z.ai, open weights): 45 · $2.01 Kimi K3, July (Moonshot AI): 44 Price per 1M tokens in / out: Kimi K3.1 Pro and GPT-6.1 Sol $2 / $10; Claude Opus 5.5 $4 / $20; GPT-6 Astra and Claude Fable 5.1 $10 / $50. To get through the index, Kimi K3.1 Pro wrote 150 million tokens; the median comparable model writes 87 million. Index and cost per task: Artificial Analysis, October 23, 2026. Prices: each maker's pricing page. Kimi K3.1 Pro's weights are promised by November 4, 2026. ⏱️ CHAPTERS 0:00 Where Kimi K3.1 Pro lands 0:36 What Moonshot shipped 1:15 Moonshot's own benchmark chart 2:00 The independent ranking 2:48 One prompt on three models 3:44 What a finished task costs 4:42 Vs the Chinese open models 5:26 How open it is right now 5:49 Which model to use (and what I use) 📚 SOURCES Moonshot AI, Kimi K3.1 Pro tech blog: https://kimi.ai/blog/kimi-k3-1-pro Moonshot AI's launch post: https://x.com/Kimi_Moonshot/status/2113680701770068067 Kimi API pricing: https://platform.kimi.ai/docs/pricing/chat Artificial Analysis, Kimi K3.1 Pro: https://artificialanalysis.ai/models/kimi-k3-1-pro Artificial Analysis's launch post: https://x.com/ArtificialAnlys/status/2113688083145314416 OpenAI, GPT-6.1 Sol and GPT-6 Astra: https://developers.openai.com/api/docs/models/gpt-6.1-sol https://developers.openai.com/api/docs/models/gpt-6-astra Anthropic pricing: https://platform.claude.com/docs/en/about-claude/pricing 🎬 CREDITS Dinosaur scenes (Kimi K3, K3.1 Pro, GPT-6.1 Sol): Peter Gostev, Arena https://www.youtube.com/watch?v=rVuwIr4wiPw Benchmark charts and launch video: Moonshot AI Intelligence Index chart: Artificial Analysis Music: Mindscape — Lennon Hutton (Epidemic Sound) #KimiK3 #Moonshot #AI