Data as of 2026-09-20.
Price spread

Qwen3.8 2.4T A95B

qwen/qwen3.8-2.4t-a95b

Qwen (Alibaba)· 991K ctx
open weightstool callstructured quant: unknown

spread (output, p90 ÷ p10)

1.0×

$6.00 – $7.50 · 5 providers · max÷min 1.3×

5 providers priced · 5 offerings · 1 omitted (stale)

Cheapest / median / spread use providers that serve the model on their own capacity: first-party lab, hyperscaler, inference host.

Provider classes in the stats:
Blended $/M assumptions

blended = 90.9% input × [(1 − 70% cache) × input + 70% cache × cache-read] + 9.1% × output + cache writes (30% of uncached input) where a write price is published

A provider that publishes no cache price is blended with cache share 0 (all input at the input price) and labelled “no cache price published” — never penalised against identical in/out prices.

ProviderInputOutputCache rdBlended
IteraCompute
unclassified
$1.95$5.95$0.20$1.20
Deep Infra
inference host
$2.00$6.00$0.20$1.22
Eden AI
router / aggregator
$2.00$6.00$0.25$1.45
Fireworks AI
inference host
$2.00$6.00$0.25$1.25
Charm Hyper
unclassified
$2.00$6.00$0.25$1.25
AIHubMix
router / aggregator
$2.00$6.00$0.50$1.41
RunInfra
unclassified
$2.00$6.00$0.20$1.22
Vercel AI Gateway
router / aggregator
$2.00$6.00$0.25$1.25
Kilo Gateway
router / aggregator
$2.00$6.00$0.25$1.45
Modal
inference host
$2.00$6.00$0.25$1.25
OpenRouter
router / aggregator
$2.00$6.00$0.25$1.25
Requesty
router / aggregator
$2.00$6.00$0.20$1.22
Cortecs
router / aggregator
$2.50$6.00$0.63$1.63
TensorX
unclassified
$2.50$6.00$0.63$1.63
Hugging Face
inference host
$2.50$6.25—$2.84no cache price published
Merge Gateway
router / aggregator
$2.50$6.25$0.50$1.57
Venice AIoutlier
inference host
$2.50$7.50$0.31$1.56
Stale offerings (not in the spread) · 1

Last priced more than 3 days before the latest snapshot — shown for reference, never mixed into cheapest, median or spread.

  • NanoGPT qwen/qwen3.8-2.4t-a95bpay per token

Data: models.dev (MIT), snapshot 2026-09-20. Prices are vendor-published; verify before purchase. Methodology.