China's Moonshot AI just dropped Kimi K3.1 Pro — the first model in its new K3.1 series — and the numbers are startling. On Terminal-Bench 4.0, one of the toughest coding-agent tests in the industry, it scores 65.7%: ahead of OpenAI's GPT-6 Astra (58.2%) and just 0.7 points behind Anthropic's Claude Opus 5.5 (66.4%). Better still, Moonshot has promised the full open weights by November 4, which would make it the strongest open-weight model in the world. Here's what you need to know.
At a glance: Kimi K3.1 Pro launched Oct 23, 2026 · 2.8T total params / 104B active (MoE) · 1M-token context · Artificial Analysis Index score of 55 (4th overall) · API 33% cheaper than K3: $2/$10 per 1M tokens · K3.1 Pro Highspeed variant runs 2–3x faster · Open weights due Nov 4, 2026.
What Is Kimi K3.1 Pro?
Kimi K3.1 Pro is not a brand-new base model. Moonshot describes it as a post-trained version of Kimi K3, pushed through far more reinforcement learning across what the company calls "hundreds of downstream domains". The scale is enormous: a mixture-of-experts architecture with roughly 2.8 trillion total parameters (104 billion active at any one time), a 1-million-token context window, and text plus image inputs.
It is live right now in the Kimi app, the Kimi Work desktop app, the Kimi Code coding tool and the Kimi API — with a companion variant, Kimi K3.1 Pro Highspeed, that Moonshot says delivers the same abilities at two to three times the speed.
Benchmarks: 0.7 Points From Claude Opus 5.5
The headline claim comes from Terminal-Bench 4.0, which tests coding agents on realistic, resource-constrained terminal tasks. Moonshot's own figures put K3.1 Pro at 65.7% — against 66.4% for Claude Opus 5.5 and 58.2% for GPT-6 Astra. Moonshot ran its model in the Claude Code harness and pulled rivals' scores from their published reports or the official leaderboard.
Elsewhere the gains are real but uneven:
- Kimi Code Bench 3.0 (long-horizon): 61.2%, up from K3's 55.95% — a solid jump in multi-step tool-using tasks.
- BenchCAD (3D reconstruction via CAD code): 92.7% with tools, behind Opus 5.5 (96.2%) and GPT-6 Astra (95.9%).
- OfficeQA Pro v2 (office-document work): Moonshot claims a state-of-the-art 64.4.
- Hallucination rate: on Moonshot's own hardest factual-accuracy test, claim-level errors fell from 23.4% (K3) to 8.3% — roughly the 65% drop the company advertises.
The benchmark figures are Moonshot's own and have not been independently verified yet. Independent confirmation comes from Artificial Analysis, whose Intelligence Index gives K3.1 Pro a score of 55 — fourth overall, behind only three closed models (Claude Opus 5.5 at 58, Claude Sonnet 5.5 and Muse Wonder Full at 56), and a full 11 points above the old Kimi K3.
It is also worth noting what the numbers don't show: Moonshot's chart has K3.1 Pro well behind GPT-6 Astra on science reasoning (Terminal-Bench Science 0.1: 38.1 vs 68.1). This is a coding-and-agents monster, not an all-rounder.
Pricing: 33% Cheaper Than K3
The API is available today at $2 per million input tokens and $10 per million output tokens (cached input at $0.20) — a third cheaper across the board than Kimi K3. The Highspeed variant costs $4/$20. For developers calling Kimi APIs, that's a straight price cut on top of the capability jump.
Open Weights by November 4 — Why That Matters
This is the part that makes K3.1 Pro more than a routine upgrade. Moonshot has committed to publishing the full model weights by November 4. Once released, Artificial Analysis' data suggests it would be the highest-scoring open-weight model on the Intelligence Index — ahead of everything else you can download and run yourself.
That puts genuine frontier-class coding and agent capability into the hands of anyone with enough hardware, free of per-token pricing. For the open-source AI ecosystem, this is a Big Moment™ — the same category of shock the DeepSeek release delivered in early 2025.
Safety note: Moonshot says it deliberately stripped cyberattack-related data from mid-training and reinforcement learning for the open release, so the public weights "won't push the frontier of offensive hacking" — its words, not an independent audit.
How to Try It Now
- Open the Kimi app or go to kimi.ai — K3.1 Pro is selectable in chat right now.
- Developers: grab an API key from the Moonshot AI platform and call the new model ID.
- Coding users: enable it inside Kimi Code for the full agentic coding experience.
- Waiting on the weights? Mark November 4 — Moonshot will publish them for download and self-hosting.
Read the detailed breakdowns at Unite.AI ↗ and MarkTechPost ↗.
FAQ
Is Kimi K3.1 Pro free to use?
The Kimi app offers free-tier access, and the API is priced at $2/$10 per million input/output tokens. The full open weights arrive by November 4 — free to download and run yourself.
Is Kimi K3.1 Pro better than Claude Opus 5.5?
Not overall — Opus 5.5 still leads on most benchmarks, including a 66.4 vs 65.7 edge on Terminal-Bench 4.0 and a big lead on science tasks. But K3.1 Pro is remarkably close for a model that will soon be open-weight, and it beats GPT-6 Astra on coding.
When do the open weights come out?
Moonshot has committed to publishing them by November 4, 2026. The API and app versions are live now.
The Bottom Line
Moonshot just put a model within 0.7 benchmark points of the world's best coding AI — and promised to give it away free in less than two weeks. Whether you're a developer eyeing cheaper, faster agents or just watching the AI race, November 4 is the date to circle. The open-weight frontier is about to move.
Will you try Kimi K3.1 Pro, or wait for the open weights? Tell us in the comments below!
