Kimi K3.1 Pro: Moonshot's 2.8T open-weight flagship officially launches
Moonshot AI officially launched Kimi K3.1 Pro on Oct 23, a mixture-of-experts model with 2.8 trillion total / 104 billion active parameters, a 1M-token context window, 65.7% on Terminal-Bench 4.0 (ahead of GPT-6 Astra, 0.7 pts behind Claude Opus 5.5), and ranked 4th on Artificial Analysis' Intelligence Index at 55. The API is live at $2/$10 per million input/output tokens; full model weights are committed by November 4.
Why it matters
Once weights ship, it would be the highest-scoring open-weight model ever — a frontier-class coding/agent model anyone can self-host or fine-tune, priced ~50% below Opus-class APIs.
What you could build with it
An indie developer could build a cost-aware code-review agent that routes routine reviews to Kimi K3.1 Pro's API and escalates only hard cases to a closed frontier model — cutting token spend roughly in half since the API is priced well below Opus-class rates, while keeping near-frontier quality on agentic coding tasks.
Does it hold up?
Too early to judge — the API went live Oct 23 and weights ship Nov 4. Early hands-on reports are split, from 'can't tell the difference from Opus 5.5' to usage limits disappearing within a minute.
Built with Kimi K3.1 Pro
- OpenAI Unplugs the Internet, Anthropic Confronts a Rogue Agent, and Kimi K3.1 Pro Closes the Gapdev.to · Deep dive noting K3.1 Pro's 55.1 Intelligence Index score vs Opus 5.5 at 57.6, and that the open-weight gap is now ~31 days of release time.
- Kimi K3.1 and Qwen4 in Arena testsalextech.ai · Covers Lumina's side-by-side 3D helicopter generation demos comparing Kimi K3.1 with Qwen4 ahead of launch.
- 'A law firm cut off Claude and installed Kimi': China's Kimi K3.1 lands 4thailetter.ai.kr · Reports an unnamed top US law firm dropping Claude/OpenAI APIs for Kimi and DeepSeek on self-hosted GPUs, plus the 65.7% Terminal-Bench 4.0 claim.
Learn more
First spotted on article: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74Tencent Hunyuan Hy3: open 295B MoE with 21B active under Apache 2.0Tencent open-sourced Hy3, a 295B-parameter mixture-of-experts model with only 21B parameters active per token, under…models · JEV 0.73NVIDIA Isaac Lab 3.0 GA + Isaac OS open runtime + Agile ONE SN humanoidMajor Isaac platform expansion at GTC Berlin Oct 21: Isaac Lab 3.0 is generally available as the open-source robot…models · JEV 0.73StepFun Step-5-Preview — open BF16 weights releasedStepFun published the full BF16 weights for its flagship Step-5-Preview agentic model (600B-parameter sparse MoE, 27B…models · JEV 0.73NVIDIA Nemotron 3.5 Retriever collection: Embed 8B/1B, Rerank 8B, Parse 631MNVIDIA's Oct 21 collection drop for agentic retrieval: Nemotron-3.5-Embed in 8B BF16, 1B BF16 and 1B NVFP4…models · JEV 0.72ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.