Kimi K3 Launch Recap: What Moonshot Shipped on July 16
Everything confirmed about the Kimi K3 release: 2.8T parameters, 1M context, $3/$15 pricing, Elo 1547, and the July 27 open-weights promise.
· news
Every article is written from primary sources — official docs, Hugging Face, first-party announcements — with a visible source list. No recycled press releases.
Everything confirmed about the Kimi K3 release: 2.8T parameters, 1M context, $3/$15 pricing, Elo 1547, and the July 27 open-weights promise.
· news
What will it take to run Kimi K3's 2.8T MoE locally? Storage per quantization level, why MoE sparsity doesn't cut memory, and realistic setups.
· local
Kimi K3 prompt caching explained: what gets cached, the 60-80% effective savings reports, worked examples, and how to structure prompts for hits.
· api, pricing
Legit ways to use Kimi K3 free: kimi.com's free tier, router trials and what they limit — plus what 'free' actually costs in quotas and privacy.
· guide
Same $3/$15 sticker, different trade-offs: caching support, rate limits, billing and latency when buying Kimi K3 through Moonshot vs a router.
· api, pricing
Kimi K3 costs 3.2× more than K2.6 on input, 3.75× on output. Elo +732, 21% fewer output tokens, 1M context — when the upgrade pays and when it doesn't.
· compare
Kimi K3 claims wins over GPT-5.5 high on launch benchmarks at lower output prices. Where the claim holds, where it doesn't, and the cost math.
· compare
Kimi K3 ($3/$15, 1M context, 2.8T) vs DeepSeek V4 Pro (1.6T, cheaper tokens): benchmarks, cost per workload and which to pick for what.
· compare
Kimi K3 matches Claude Sonnet's $3/$15 pricing exactly. What you get for the same money: context, weights openness, caching and coding trade-offs.
· compare