Research2026-07-17

Independent testing firm Artificial Analysis rated Moonshot AI's new Kimi K3 model third on its overall intelligence ranking, on par with Anthropic's Opus 4.8 and OpenAI's GPT-5.5 but behind the two current leaders. Kimi K3 topped one benchmark for automated business workflows and cost about half as much per task as Opus 4.8, though Moonshot raised its prices sharply compared with the previous version. The model is available only through Moonshot's own API for now, with the company saying it intends to publish the weights later.

What changed

The previous Moonshot model, K2.6, scored 13 points lower on the index and its output tokens were priced at $4 per million.

What it unlocks

Access to third-place-tier model quality on agent-style and long-horizon knowledge work at roughly half the per-task cost of Claude Opus 4.8.

  • 57 on the Artificial Analysis Intelligence Index, #3 overall
  • $3.00/$15.00 per 1M input/output tokens, cached input $0.30
  • $0.94 average cost per task vs $1.80 for Opus 4.8
  • 2.8T total parameters, 1M context window
  • 21% fewer output tokens than K2.6 (132M vs 166M)
  • hallucination rate rose from 39% to 51%

What you need to act on it

  • access to Moonshot AI's first-party API
  • self-hosting not yet possible as weights are unreleased

Send this to someone who needs it

Shares the story and its sources. Nothing about you.

What does this mean for your job?

This is the story as everyone gets it. Once a week we send you the version written for your role — what changed, why it matters for the work you actually do, and one thing to try. Free while we tune it.