THE SCOREBOARD FOR WHAT'S NEXTAI + QUANTUM, SCORED. COMPARED. EXPLAINED.
AIQUANTUMSCORE.
← AI & Quantum News

AI / NEWS BRIEF

Moonshot launches Kimi K3.1 Pro ahead of an open-weight release

Kimi K3.1 Pro is live through Moonshot's apps and API, with full weights promised by November 4. Independent testing places it near the leading proprietary models.

By AIQuantumScore · · Event 2026-10-23

Hosted access now, weights later

Moonshot has released Kimi K3.1 Pro and a Highspeed variant through Kimi.ai, Kimi Work, Kimi Code and its API. The company says the model retains a one-million-token context window, while Highspeed generates responses two to three times faster. Full model weights are not yet available: Moonshot says it will publish them by November 4. Artificial Analysis lists first-party pricing at $2 per million input tokens, $10 per million output tokens and $0.20 per million cache-hit tokens.

Independent results support a major capability gain

Artificial Analysis scored K3.1 Pro at 55 on version 4.3.2 of its Intelligence Index, fourth among the models it compared and 11 points above Kimi K3. Its evaluation measured 67% on Terminal-Bench 4.0, 73% on AutomationBench-AA and 1,846 Elo on GDPval-AA. Those results independently support a substantial agentic-performance improvement, but remain specific to Artificial Analysis's methodology and should not be treated as a universal ranking or as a REVAL score.

Cyber capability was deliberately reduced

Moonshot says it removed cyber-related data from parts of training and reinforcement learning before the planned open-weight release. On the company's internal xdev-bench-hard test, the final model scored 81.0% compared with 90.5% for a checkpoint without those risk-reduction steps. Reuters notes that these cyber results have not been independently verified. The model still substantially exceeded its predecessor on Moonshot's test, illustrating the difficulty of reducing offensive capability while preserving broad coding gains.

Why it matters

OUR ANALYSIS

If Moonshot publishes the promised weights, K3.1 Pro would become the highest-scoring open-weight model yet measured by Artificial Analysis and narrow its index gap with leading proprietary systems. That could materially expand access to frontier-level agentic models, although the 2.8-trillion-parameter architecture will remain costly to self-host. The launch should be tracked using source-labelled external observations; it does not justify inventing an AIQuantumScore or REVAL result before controlled testing.

More to follow