Key points
- Moonshot AI released Kimi K3.1 Pro, scoring 55 on the Artificial Analysis Intelligence Index, fourth overall behind Claude Opus 5.5 (58), Claude Sonnet 5.5 and Muse Wonder Full (both 56).
- API prices fell by one-third: $2 per million input tokens and $10 output, down from $3 and $15. A faster Highspeed variant costs twice as much, per Latent Space.
- The model is live in Kimi's app, Kimi Work, Kimi Code and the API; Moonshot promises open weights by November 4, 2026.
- Context window stays at 1 million tokens, with 2.8 trillion total and 104 billion active parameters, the same as K3.
Why it matters for marketers: Near-frontier agentic and coding performance at lower list prices gives teams a cheaper model to test for workflows, though cost per finished task may differ from token rates.
Moonshot AI has launched Kimi K3.1 Pro, a post-trained update to its Kimi K3 model that uses far more post-training compute but keeps the same size. Artificial Analysis ranks it fourth on its Intelligence Index with 55 points, an 11-point jump over K3. The biggest gains are in agentic tasks, where it comes close to Claude Opus 5.5 on the GDPval-AA benchmark.
The model is available now through Kimi's website and app, Kimi Work, Kimi Code and the API. Moonshot also shipped a Highspeed version it says runs two to three times faster. It added a 3D Make feature on Kimi.com and, according to Latent Space, Kimi Hosted Agents for long-running work.
For marketers and content teams, the draw is price and agent capability. Input and output tokens cost a third less than K3. But Artificial Analysis found K3.1 Pro uses about 14 percent more output tokens per task, so its cost per Intelligence Index task is $1.98, nearly the same as K3's $2.00. Moonshot's own claims, such as lower hallucination rates and finance gains, come from internal benchmarks and are not independent results.
Watch the weights. Moonshot says it will publish them by November 4, 2026, and Artificial Analysis says that would narrow the gap between open and proprietary models from 12 points to 3. The license and hardware needs are not yet known, and a 2.8 trillion parameter model is not light to self-host. Teams can test the API now against their own tasks and measure cost per finished job rather than per token.
Sources
This brief is our own summary. Read the full reporting at the source.
- The Information: Moonshot AI Launches New Kimi K3.1 Pro Model
- The Decoder: Chinese AI model Kimi K3.1 Pro is closing in on Anthropic's Claude Opus 5.5
- Latent Space: [AINews] Kimi 3.1 Pro 104BA-2.8T: Fable-5.1-Class Open Weights narrows Closed-AI lead to 1 month... and Pro Highspeed API with 2-3× speed
- MarkTechPost: Moonshot AI Releases Kimi K3.1 Pro: A 2.8T Total/104B Active Params (MoE) Model with 1M Context
