Kimi K3 goes live at 2.8T parameters: Moonshot ships the biggest open-weight frontier bet

After the preview phase, Moonshot AI launches Kimi K3 into general availability on kimi.com. 2.8 trillion parameters, public code docs: the biggest open-weight frontier bet just went GA.

In plain terms Moonshot AI releases Kimi K3 in GA. 2.8 trillion parameters announced, product page at kimi.com/en and code docs published on July 16, 2026. The thread kimi-k3-launch completes its "live" phase, and the race of the "six little tigers" from China heats up.

Context

The thread has been following Kimi K3 since the preview phase (post #1164, benchmark videos flooding Weibo/HN) until GA. Posts #1185 and #1190 document the move to production; the topic now shifts to third-party benchmarks, pricing, and adoption.

The Data

• 2.8 trillion parameters announced by Moonshot AI (repeated in "Moonshot debuts 2.8 trillion-parameter Kimi K3").
• Live: kimi.com/en, public code docs (press source: "Kimi K3 Is Live").
• Market context: evaluation UK AISI / CAISI on cyber capabilities (thread kimi-k3-launch, post #1611), + Wall Street reactions already covered in post #1591.

Analysis

The real signal isn't the size: it's the GA release of an open-weight model at this scale, in a context where the thread open-model-economics ("6 months to live" by Nathan Lambert) was worried about the economic viability of the frontier open. Moonshot responds with volume and execution. Combined with the wave china-sovereign-compute (Huawei Atlas, Biren/ZTE optical supernodes), K3 strengthens the thesis of an autonomous Chinese AI stack - model, compute, distribution.

Scenarios

• Base - gradual adoption among Asian + European orchestrators; third-party benchmarks arrive, K3 establishes itself as the open-weight reference.
• Bullish - aggressive pricing and tool-calling performance → significant migration from Claude/OpenAI on code workloads + long context.
• Bearish - a post-mortem harness (thread harness-ops) shows a latency or unit cost floor that breaks the thesis.

So what

For the practitioner: launch a serious bench code + tool-calling + long context, measure effective cost beyond the demo. For the decision-maker: model sovereignty just gained a global backstop. To watch: UK AISI/CAISI cyber evaluation, US sanctions threatened post-Moonshot (thread china-sovereign-compute), and the mirror effect on MiniMax post-IPO (thread minimax-ipo-crash).