Kimi K3 isn't Mythos-class. It's something stranger.

What actually shipped

Moonshot unveiled Kimi K3 on July 16 — "the world's first open 3T-class model." Live on their API today at $3/M in, $15/M out; weights promised on Hugging Face by July 27, technical report to follow. Until that upload happens, "open-weight" is a press release, not an artifact — worth remembering for the next eleven days. One checkpoint, one reasoning tier ("max" — lower tiers "planned"), 1M-token context, vision input.

And the parameter count is either 2.8T (Moonshot) or 2.7T (Fortune), depending on who you read — the active count is undisclosed entirely. We're one day in and the most basic spec is already ambiguous. Benchmarks-vs-reality blogging remains a growth industry.

The numbers, sorted by who measured them

House rule since the SWE-Bench Pro funeral: vendor numbers and independent numbers don't get to sit in the same column.

Independent (Artificial Analysis, Arena):

An open-weight model at #1 on a blind developer-preference arena, above Fable 5, is a first. It's one arena, frontend-only — but it's a real one.

Vendor (Moonshot's launch table — which, credit where due, footnotes its own harness-mixing): DeepSWE 67.5 vs Fable's 70.0 and Sol's 73.0 (the 67.3 on Datacurve's own leaderboard roughly confirms it); Terminal-Bench 88.3 on their harness vs the 85.0 AA measured — a 3.3-point harness delta, exactly the confound post 003 was about. Ofir Press flags that their Program Bench number uses average-implementation-percentage, which flatters partial work. Standard launch-chart physics apply.

Is it Mythos-class? No — and Moonshot agrees

Read the launch post's limitations section, because it's the most honest text any lab shipped this year: "K3 nonetheless exhibits a noticeable gap in user experience compared with Claude Fable 5 and GPT 5.6 Sol." They flag sensitivity to thinking history (quality degrades if your harness drops reasoning context, or you switch models mid-session) and excessive proactiveness (on ambiguous instructions it may just… act — constrain it in AGENTS.md). Those are precisely the long-horizon reliability properties that separate a Mythos-class orchestrator from a very strong executor, and Moonshot named them unprompted.

The independent picture agrees: 2.8 points below Fable on the composite, a knowledge-reliability score half of Grok's, and a hallucination rate that AA measured regressing to 51%, from K2.6's 39%. The people who tested it land the same way. Ethan Mollick: "Very good model, not Sol Max or Fable, but great open weights… closest to the frontier yet." Simon Willison found the vision input solid and the pelican acceptable — at 25 cents a pelican, 13,241 reasoning tokens included. MIT/DeepMind's Michiel Bakker, on the capability jump: "These results seem impossible to explain through distillation alone." And OpenAI's own Dean Ball, after hands-on agentic sessions: "very good," matching "the best public models from Q1 2026" — five words of praise carrying six months of asterisk.

The most honest sentence any lab shipped this year is Moonshot admitting their own model isn't the frontier. The scoreboard agrees. The scoreboard also says: barely.

How they got here: the technique

K3's launch post reads less like marketing than an architecture changelog, and the architecture is where this release actually matters:

• Extreme sparsity. 896 experts, 16 active per token — roughly 1.8%. For scale: K2.6 ran 1T total / 32B active; DeepSeek V4 Pro 1.6T/49B; GLM-5.2 753B/40B. K3 nearly doubles the biggest open total ever while activating a sliver of it. (Active count undisclosed — so per-token compute comparisons are guesswork until the tech report.)
• Kimi Delta Attention (KDA) — a hybrid linear-attention mechanism for information flow across the 1M context, plus Attention Residuals, which retrieve representations selectively across depth instead of accumulating uniformly.
• Quantile Balancing — expert routing balanced directly from router-score quantiles, deleting the usual pile of load-balancing hyperparameters.
• Per-Head Muon — the Muon optimizer extended to optimize each attention head independently — and a SiTU activation, plus quantization-aware training from SFT onward (MXFP4 weights / MXFP8 activations).
• Net claim: ~2.5× scaling efficiency over K2. Vendor number, unverified, tech report pending — but Emad Mostaque's estimate of a $15–25M training cost at ~1e25 FLOPs is the kind of figure that, if even directionally right, explains why "Scale isn't all you need!" was his conclusion.

The bill: this thing is slow

Architecture giveth, architecture taketh. AA measured 62 tokens/second (peer median ~72.7), ~2 seconds to first streamed chunk but 32 more seconds of reasoning before the first answer token, ~42 seconds end-to-end for a 500-token response. And it consumed 130M output tokens to complete the Intelligence Index against a 63M peer median — double appetite. Three compounding causes: only "max" reasoning effort exists at launch, so every request pays the full thinking tax; the model is verbose by temperament; and the 896-expert layout is why Moonshot recommends 64+ accelerator supernodes for serving — this architecture wants a data center, and even then it saunters. The one mercy: at $3/$15 per million, the full AA index cost $2,691 on K3 vs $2,824 on Sol and $5,631 on Fable. Cheap per token, hungry in tokens, slow in wall-clock — pick the two you can live with.

What to actually do

• Pilot it now, via API, on frontend and agentic coding — the Arena #1 and Terminal-Bench 85 are real signals, and per-task cost is Sol-tier or better.
• Don't orchestrate with it yet. Thinking-history sensitivity + excessive proactiveness + 51% hallucination rate is the wrong profile for the top of an agent tree. Keep a Mythos-class model in that seat; give K3 the execution lanes.
• Don't put it in anything latency-facing. 34 seconds to first answer token is a batch-job number.
• Wait for July 27 before repeating "open-weight" — and for the tech report before repeating "2.5× efficiency."
• Recalibrate your priors: the gap between the best open model and the best model, full stop, is now 2.8 points on the only independent composite we have. Twelve months ago it was a chasm. That fact is bigger than any single row in the table — and it's what part 2 is about.