Inkling inkling

thinkingmachines · thinkingmachines

#4 overall

params
975B
arch
moe
context
512k
license
apache-2.0 open weights
released
Jul 2026
train compute
1.1e+25 FLOP
downloads/30d
2.1M

# architecture

attn shared GQA 64q/8kv router top-6 of 256 +2 shared expert ×256 expert out ×66 layers ctx 524,288 975B pool
layers
66
d_model
6144
heads
64q / 8kv
head dim
128
experts
top-6 of 256 + 2 shared
vocab
201024
family
inkling_mm_model

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: #4 — score 13.3 ● high — all signals present

signalweightinput (0–100)
aa_intelligence 0.70 7.7
bench_composite 0.30 26.4

benchmark panel evidence:

complete panelscore (0–100)
aime-2026-v126.4

full methodology

# trend

OGM score · last 99 days
OGM score over 99 days
overall rank (up is better)
Overall rank over 99 days

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

# benchmarks

benchmarkscoresourcedate
AA Intelligence Index 12.5 Artificial Analysis —
Aa-omniscience-index 2.1 / 200 LLM Stats —
AIME 2026 1.0 / 1 LLM Stats —
Charxiv-r 0.8 / 1 LLM Stats —
Gdpval-aa 1238.0 / 3000 LLM Stats —

# where to run

providerquantctx$/M in$/M out$/M cacheprice srctpsuptime✓
DeepInfra 512k $0.95$4.05— via hfrouter 169 —
BaseTen 1024k $1.00$4.05— via litellm — —
Fireworks 1024k $1.00$4.05$0.17 via modelsdev — —
Togetherunknown 512k $1.00$4.05$0.17 via openrouter — 98.4%

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
inkling
epoch
Inkling
openrouter
thinkingmachines/inkling