Compare Qwen3.8 27B API routes, input/output and cache prices. Calculate monthly token costs, inspect provider conditions and export your budget.
Source-linked prices · updated 9 Nov 2026. Text-token usage rates in USD. Funding fees, taxes and separately billed tools are excluded. Default routing, service tiers and long-context conditions are attached to each quote.
How much will Qwen3.8 27B cost for my token mix?
Use the input, output and actual cache-read prices from one route, then add the funding and other charges that apply to your account. A low input-token price alone does not establish the lowest workload cost. The examples below keep each route’s prices together and the calculator lets you replace the token volumes.
This page covers 30 collected route quotes from Crusoe and DeepInfra and NanoGPT and Nebius Token Factory and Novita AI and OpenRouter. Prices were observed 9 Nov 2026. Each route’s context, precision and pricing conditions remain attached below. Enter your token mix →
Worked token budgets
Calculated examples for selected routes, not performance measurements or a ranking. All amounts use one route’s paired base rates. The cache example assumes actual hits; funding, cache creation, tax and separately billed extras are excluded.
Light use: 1M input + 0.2M output. Output-heavy: 10M input + 10M output. Cache example: 10M input, including 8M cache reads, + 2M output. These can represent a month or another planning period; individual request limits still apply.
What will your token mix cost?
Choose a model listing and enter your total token volume for any planning period. Cache reads are included within input tokens. Routes are grouped by the channel’s model identity.
Base text-token estimates. Check long-context thresholds, cache creation, modality and service-tier pricing in each route’s evidence. Enter request count only to include a published per-request fee; missing required rates produce an incomplete estimate. Prices do not establish equivalent performance or served weights.
Enable JavaScript to calculate your workload. Published rates remain available below.
Route
Base usage · USD
Cost breakdown
Endpoint quotes 30
30 route quotes
AkashMLvia OpenRouter
Qwen: Qwen3.8 27B
Conditional quote
Input / 1M tokens
$0.225
Output / 1M tokens
$1.98
Cache read / 1M tokens
$0.05
Cache write / 1M tokens
Not published
fp8 · Tier not published · 262,144 context tokens · 131,072 max output tokens
Third-party base text-token rate via GPU Rental Prices (CC BY 4.0). Converted from USD/million tokens. Long-context, cache creation, modality, platform, tax and other charges are not fully covered; check provider conditions.
Latest collection attempt
success
Darkbloomvia OpenRouter
Qwen: Qwen3.8 27B
Conditional quote
Input / 1M tokens
$0.035
Output / 1M tokens
$1.5
Cache read / 1M tokens
$0.0175
Cache write / 1M tokens
Not published
fp4 · Tier not published · 262,144 context tokens · 65,536 max output tokens
Third-party base text-token rate via GPU Rental Prices (CC BY 4.0). Converted from USD/million tokens. Long-context, cache creation, modality, platform, tax and other charges are not fully covered; check provider conditions.
Latest collection attempt
success
DeepInfravia OpenRouter
Qwen: Qwen3.8 27B
Conditional quote
Input / 1M tokens
$0.15
Output / 1M tokens
$1.875
Cache read / 1M tokens
$0.0375
Cache write / 1M tokens
Not published
bf16 · Tier not published · 262,144 context tokens · 235,929 max output tokens
Third-party base text-token rate via GPU Rental Prices (CC BY 4.0). Converted from USD/million tokens. Long-context, cache creation, modality, platform, tax and other charges are not fully covered; check provider conditions.
Latest collection attempt
success
Novitavia OpenRouter
Qwen: Qwen3.8 27B
Conditional quote
Input / 1M tokens
$0.42
Output / 1M tokens
$3
Cache read / 1M tokens
$0.085
Cache write / 1M tokens
Not published
unknown · Tier not published · 1,000,000 context tokens · 131,072 max output tokens
Third-party base text-token rate via GPU Rental Prices (CC BY 4.0). Converted from USD/million tokens. Long-context, cache creation, modality, platform, tax and other charges are not fully covered; check provider conditions.
Latest collection attempt
success
Parasailvia OpenRouter
Qwen: Qwen3.8 27B
Conditional quote
Input / 1M tokens
$0.24
Output / 1M tokens
$2.2
Cache read / 1M tokens
$0.05
Cache write / 1M tokens
Not published
fp8 · Tier not published · 262,144 context tokens · 235,929 max output tokens
Standard pay-as-you-go text-token base rates; default route, no provider pinned. Long-context, video, service tiers, deposits, tools and tax require separate checks.
Additional pricing conditions
{"video_request":{"note":"Applies to all tokens in requests containing video, including subsequent turns that retain the video.","prompt":"0.32","completion":"2.5","cached_prompt":"0.032"}}
Latest collection attempt
success
No matching endpoints. Try a model, channel or provider name, or clear the filter.
Check the billing behind the quote.
Why does the same model have several API prices?
OpenRouter sells access through named upstream routes. Precision, context limits, output limits and service tiers belong to each route, so inspect those fields alongside the rates. Multiple endpoint records are not necessarily multiple independent companies.
NanoGPT’s record here is its default pay-as-you-go route. A list of possible upstream providers does not make that price a quote for each pinned provider. Its default route and an OpenRouter endpoint also do not establish that both serve identical weights or deliver the same quality and speed.
A cache-read rate prices tokens that were actually read from cache. Repeated text, a cache-capable provider or a successful request is not proof of a hit. Enter the cache-read percentage from your usage records; leave it at zero for an uncached budget.
When a route has no published cache-read rate, an estimate with cache reads is incomplete. The calculator does not replace the missing value with zero. Cache creation, expiration and changes in the route may add costs that this base-token estimate does not predict.
Do these prices cover video, long contexts and every tier?
The calculations use the base text-token prices in each quote. The NanoGPT Qwen3.8 27B listing includes separate video-request pricing; inspect that condition before applying the text rates to video-containing requests. Long-context overrides and service tiers must also be checked on the route you will use.
An aggregate of ten million input tokens describes a budget period, not a single request. It does not prove that any individual request fits the endpoint’s context or output limits. Check the largest request, modality and supported parameters separately.
Can I treat the lowest input price as the cheapest Qwen3.8 27B API?
Calculate both input and output for your workload, including cache reads and applicable fees. Then check the route’s precision, context, output and service conditions. These quotes alone do not establish a quality-adjusted cost winner.
Are these NanoGPT subscription prices?
No. The NanoGPT quote uses default pay-as-you-go base rates. Subscription eligibility, limits and pinned-provider charges are separate conditions. The related channel-cost guide explains those distinctions.
Does this show that Qwen3.8 27B is cheaper to self-host?
No deployment performance is measured here. Use the API budget as one input to the API-versus-hardware guide, then supply task-equivalent runtime, memory, power and operating costs for your proposed hardware.
