#kimi k3

A worked agency research workflow on Kimi K3 Swarm, where the merge-and-dedup step, not the fan-out, decides whether the output is worth reading at all.

Moonshot trained Agent Swarm against two failure modes it names itself, serial collapse and fake parallelism. Use them as your own fan-out decision test.

K3's least-used surfaces are its 1M context, native vision and reasoning-effort dial. Switching effort mid-session forfeits the cache rate that funds it.

Kimi K3 leads open-weights models on intelligence, but its hallucination rate climbed to 51%. Why you must run your own evals before adopting it.

Kimi K3's weights are open, but the licence is bespoke: a $20M revenue trigger for resellers, UI attribution rules, and a 1.56TB download to plan for.

Kimi K3 is 1.56TB of weights and needs 64+ accelerators to serve. Here is the real cost maths of self-hosting a trillion-parameter model in 2026.

Between July 17 and 23, seven notable models shipped from five vendors — most efficiency or positioning plays, not capability jumps. A triage framework.

The White House OSTP accused Moonshot of distilling Anthropic's Fable 5 to build Kimi K3. Treasury floated sanctions; nothing is enacted so far.

Gemini 3.6 Flash matches its predecessor on Artificial Analysis intelligence but halves time per task; recomputed list prices make it the cheaper default pick.

Running Kimi K3 in Kimi Code: the Moderato plan unlocks 256K context, Allegretto the full 1M, and cache discipline — not knobs — controls your real cost.

Kimi K3's open weights are promised by July 27, not shipped — use the 10-day window to prep hosting for a 2.8T model and check the license before you commit.

Kimi K3's open-weight bet meets Anthropic's closed Fable 5. Vendor benchmarks split 6-8, K3 lists at $3/$15 vs $10/$50, and weights are promised July 27.

Kimi K3 leads GPT-5.6 Sol on seven of fourteen vendor-reported benchmarks, but Sol's effort controls and ultra mode reframe agentic fit beyond raw scores.

Five open-weight moves cluster in one July window — K3, Inkling, M3 Pro, a Mistral MoE teaser and a scheduled DeepSeek V4 — narrowing the gap to a generation.

Kimi K3 shipped OpenAI-SDK compatible on day one — the fifth line in 2026. 'Compatible' covers chat and tools but leaks on reasoning deltas and sampling.

Kimi K3 brings 2.8T parameters, 1M context, and near-frontier vendor benchmarks, with open weights due July 27, 2026. What the release means for AI buyers.