Kimi K3 kimi-k3

moonshotai · kimi

#2 overall

params
2800B
arch
moe
context
1024k
license
other open weights
released
Jul 2026
train compute
2.0e+25 FLOP
downloads/30d
874.5k

# architecture

attn shared MHA ×96 router top-16 of 896 +2 shared expert ×896 expert out ×93 layers ctx 1,048,576 2800B pool
layers
93
d_model
7168
heads
96q / 96kv
head dim
74
experts
top-16 of 896 + 2 shared
vocab
163840
family
kimi_linear

signal path of one layer, generated from the registry's structured fields — dimension lines quote the model's real numbers; an MoE trace forks at the router, a dense trace runs straight through. Geometry fields come from the repo's config.json.

# why ranked

Overall: #2 — score 98.3 ● high — all signals present

signalweightinput (0–100)
aa_intelligence 0.70 97.6
bench_composite 0.30 100.0

benchmark panel evidence:

complete panelscore (0–100)
aa-coding-v1100.0
browsecomp-v1100.0

Coding: provisional — score 100.0 ● low — a single signal; treat with caution

provisional: too few independent signals for a numbered position — the score is shown, but this model sorts after every ranked model.

signalweightinput (0–100)
aa_coding 0.30 100.0
aider_polyglot 0.30 —
swe_bench_verified 0.40 —

missing signals are dropped and the remaining weights renormalized — never imputed.

full methodology

# trend

OGM score · last 87 days
OGM score over 87 days
overall rank (up is better)
Overall rank over 87 days

methodology changed during this history window; score movement across that boundary is not model movement. See methodology v6.

# benchmarks

benchmarkscoresourcedate
AA Coding Index 76.2 Artificial Analysis —
AA Intelligence Index 43.6 Artificial Analysis —
Aa-briefcase 1548.0 / 3000 LLM Stats —
Apex-agents 0.4 / 1 LLM Stats —
Automationbench 0.3 / 1 LLM Stats —
Babyvision 0.9 / 1 LLM Stats —
Browsecomp 0.9 / 1 LLM Stats —
Charxiv-r 0.9 / 1 LLM Stats —
Mathvision 1.0 / 1 LLM Stats —

# where to run

providerquantctx$/M in$/M out$/M cacheprice srctpsuptime✓
Featherless — ——— via hfrouter — —
Nebius — ——— nebius — — ✓
SiliconFlow — ——— siliconflow — — ✓
Sail Researchfp4 1024k $1.20$5.50$0.20 via openrouter — 100.0%
Sail Researchfp4 1024k $1.20$5.50$0.20 via openrouter — 100.0%
Relacefp4 1024k $0.17$9.75$0.17 via openrouter — 100.0%
InferenceNetfp4 1024k $0.24$10.80$0.21 via openrouter — 100.0%
InferenceNetfp4 1024k $0.24$11.94$0.21 via openrouter — 100.0%
Wafermxfp4 1024k $0.36$11.90$0.36 via openrouter — 100.0%
Decartmxfp4 1024k $1.65$8.25$0.17 via openrouter — 100.0%
InferenceNetfp4 1024k $0.50$11.88$0.35 via openrouter — 100.0%
Chutesmxfp4 1024k $1.85$9.50— chutes — 50.0% ✓
Wafermxfp4 1024k $0.17$16.00$0.16 via openrouter — 100.0%
Fireworksunknown 1024k $2.10$10.50$0.21 via openrouter — 94.0%
Makorafp4 1024k $0.65$15.00$0.23 via openrouter — 100.0%
AkashMLfp4 1024k $1.20$14.00$1.20 via openrouter — 100.0%
Morphfp8 1024k $1.15$16.09$0.41 via openrouter — 99.9%
Sail 1024k $2.50$12.50— via litellm — —
DigitalOceanunknown 1024k $2.55$12.95$0.26 via openrouter — 100.0%
InferenceNetfp4 1024k $2.24$14.52$0.35 via openrouter — 100.0%
Scx 1000k $2.83$14.13— via litellm — —
DeepInframxfp4 1024k $2.85$14.25— deepinfra — 100.0% ✓
Aihubmix 1024k $3.00$15.00— via litellm — —
AtlasCloudint4 1024k $3.00$15.00— atlascloud — — ✓
BaseTenfp8 1024k $3.00$15.00$0.30 via openrouter — 100.0%
Fireworksunknown 1024k $3.00$15.00$0.30 via openrouter — 93.4%
Modalmxfp4 1024k $3.00$15.00$0.30 via openrouter — 100.0%
Moonshot AImxfp4 1024k $3.00$15.00$0.30 via openrouter — 100.0%
Novita 1024k $3.00$15.00— novita — — ✓
Opencode Go 1024k $3.00$15.00— via litellm — —
Parasailfp4 1024k $3.00$15.00$0.30 via openrouter — 98.3%
ParetoAlphaunknown 1024k $3.00$15.00$0.30 via openrouter — 71.4%
Phalaunknown 1024k $3.00$15.00— phala — 100.0% ✓
Tencent 1000k $3.00$15.00— via litellm — —
Togetherunknown 1024k $3.00$15.00$0.30 via openrouter — 99.9%
Amazon Bedrockunknown 1024k $3.30$16.50— bedrock — 100.0% ✓
Alibabaunknown 1024k $3.45$17.25$0.34 via openrouter — 100.0%
Venicenot-available 1000k $3.75$18.75— venice — — ✓
Fireworksunknown 1024k $4.50$22.50$0.45 via openrouter — 90.0%
Fireworksunknown 1024k $4.50$22.50$0.45 via openrouter — 99.3%

sorted by blended price ((3·input + output) / 4 per 1M) · ✓ = the provider's own catalog confirms the offer · "via …" prices are what the aggregator routing the offer charges, not the provider's own list price

source aliases
aa
kimi-k3
epoch
Kimi K3
llmstats
kimi-k3
openrouter
moonshotai/kimi-k3