Moonshot released its latest reasoning model, Kimi K3.1 Pro, on 2026‑10‑23. The model placed fourth on the Artificial Analysis Intelligence Index with a score of 55. It trails only Claude Opus 5.5 (58), Claude Sonnet 5.5 (56) and Muse Wonder Full (56). The company says the model will become the highest‑scoring open‑weights model once the weights are published.
How Kimi K3.1 Pro performs
The biggest gains are in agentic benchmarks, which test a model’s ability to run software tools. K3.1 Pro scores 1,846 Elo on GDPval‑AA, 312 points higher than the previous K3, and ranks third behind Claude Opus 5.5 (1,866) and Muse Wonder Full (1,857). It also leads Terminal‑Bench 4.0 and improves its AutomationBench‑AA score from 58% to 73%. Other tests show smaller jumps: Humanity’s Last Exam (47% → 52%), CritPt (23% → 25%) and SciCode (59% → 61%). The model’s hallucination rate fell from 53% to 44% while overall accuracy rose from 48% to 51%.
Pricing and technical specs
Moonshot lists the cost per task at about $1.98, roughly the same as the earlier K3 model. The new model uses 55 K output tokens per Intelligence Index task, 14% more than K3, but its token prices are 33% lower across input, output and cache‑hit rates. K3.1 Pro supports a 1 million‑token context window, runs on 2.8 trillion total parameters with 104 billion active ones, and accepts text and image inputs. Weights are not yet published but are expected soon, which would make it the top open‑weights model on the Index.
Why it matters
For developers who need AI that can run software tools, Kimi K3.1 Pro offers noticeably better performance on agentic tasks without raising costs. When Moonshot releases the model’s weights, open‑source users will have a high‑scoring alternative to proprietary models, narrowing the gap in capability.
