sao10k/llama-3-8b-stheno-v3.2
○ params○ context○ cutoff○ released○ licence
○
sao10k/llama-3-8b-stheno-v3.2 costs $0.05 / $0.05 per 1M tokens in/out at the cheapest of 1 seller.
Cheapest: Novita AI · read 2026-10-22 ↗
Hosted pricing — 1 seller, cheapest first
The cheapest hosted rate is $0.05 in / $0.05 out at Novita AI. Every row cites the pricing page it was read from.
SellerInput ↓OutputCached inBatchRead
Cost calculator
INPUT TOKENS / MO120M
OUTPUT TOKENS / MO40M
CACHED SHARE OF INPUT0%
Sellers without a cached tier bill it at full input rate
Novita AI$8 /mo
Self-hosting — what it takes to run
QuantMemoryCheapest GPU that fitsEst $/mo 24/7
4-bit○○○
8-bit○○○
16-bit○○○
memory needs the model's parameter count and weights size · no seller publishes either to us · rates live from /gpu
GPUs that can run this model
Live cheapest rates · from /gpuB300-SXM-288GB →○ · ~288 GB$4.00/hrGB300-SXM-288GB →○ · ~288 GB$12.67/hrMI325X-256GB →○ · ~256 GB$3.80/hrB200-SXM-192GB →○ · ~192 GB$3.74/hrMI300X-192GB →○ · ~192 GB$2.39/hr
each row links to that GPU’s live price table · whether this model fits is not computed, because weights and KV cache are figures we do not hold
Capabilities
Function calling○ no answer on record
Structured JSON output○ no answer on record
Vision input○ no answer on record
Audio input○ no answer on record
Fine-tuning API○ no answer on record
System prompt○ no answer on record
Benchmarks
⚑ Creator-reported — not independently reproducedMMLU-Pro○
GPQA Diamond○
HumanEval○
MATH-500○
MMMU (vision)○
SWE-bench○
A ring means nobody has told us · we run no evaluations of our own
Vs its class
In-category only · other Sao10k models we price
ModelIn/Out $/1MContextFunction calls
Often compared against
Which other models readers put beside this one on a compare page · last 30 days
Nobody has compared this model against another in the last 30 days.
Frequently asked
What does sao10k/llama-3-8b-stheno-v3.2 cost per million tokens?
From $0.05 input / $0.05 output at Novita AI, the cheapest of 1 seller.
What's the cheapest way to run it?
Hosted at Novita AI on today's rates. Whether self-hosting beats that above some volume is a calculation we cannot finish: it needs the weights size at a stated precision, and no seller publishes one to us.
How current are these prices?
Read from each seller's own page or API, most recently on 2026-10-22. Anything older than 21 days is hidden rather than shown stale, so a missing seller is never an estimate.
Is sao10k/llama-3-8b-stheno-v3.2 free for commercial use?
We hold no licence for this model. A licence belongs to the weights rather than to the endpoint, and the sellers we read publish prices rather than terms — so the honest answer is to check the model card on the maker’s own site.
Does it fit on one H100?
We cannot say. It needs the weights size at a stated precision plus KV cache at your context length, and we hold neither figure. The GPUs listed above carry their real memory and their live rate, which is the part of the question we can answer.
Does it support function calling and JSON output?
No answer on record. We read prices from sellers, not capability matrices, and a capability nobody stated is not the same as one that is absent.
Cite this page
infratoapp, “sao10k/llama-3-8b-stheno-v3.2 — hosted pricing across 1 seller”, read 22 October 2026. $0.05 input / $0.05 output per million tokens at Novita AI. https://www.infratoapp.com/llm-prices/sao10k--llama-3-8b-stheno-v3.2
Reading date travels inside the citation. How we read this data →