# Nscale

serves 12 tracked open-weights models.

models
12
verified
—
of overall top 10
—
cheapest on
4 models
vs best $
1.17×
ctx (median)
128k
tps (median)
81
ttft (median)
400ms

# models

model rank $/M in $/M out src ✓
glm-5.3 — $1.40 $4.40 hfrouter —
gpt-oss-120b — $0.10 $0.40 hfrouter —
gpt-oss-20b — $0.05 $0.20 hfrouter —
llama-3.3-70b-instruct — $0.20 $0.20 litellm —
llama-4-scout — $0.09 $0.29 hfrouter —
qwen2.5-coder-32b — $0.06 $0.20 hfrouter —
qwen3-14b — $0.07 $0.20 hfrouter —
qwen3-235b-a22b — $0.20 $0.60 hfrouter —
qwen3-235b-a22b-instruct-2507 — $0.20 $0.60 hfrouter —
qwen3-32b — $0.08 $0.25 hfrouter —
qwen3.8-27b — $0.20 $2.50 hfrouter —
qwq-32b — $0.18 $0.20 litellm —

all providers · machine-readable: providers.json