Guides / qwen3.6 27b pricing
Qwen3.6 27B pricing, and a costly pairing
Qwen3.6 27B pricing on Groq is $0.60 per million input tokens and $3.00 per million output tokens, a five-to-one spread, and the price book records no cached rate for it. A model tuned to reason out loud on a five-times output spread with no cache is the most expensive combination here. Culpa, a local-first LLM cost, margin, and forecast ledger, meters output tokens per feature so the reasoning is priced.
Why this happens
Two facts about this model compound rather than add. The output spread is five to one, which is frontier territory rather than open-weight territory, so anything that lengthens a response is expensive. And there's no cached rate to soften the input side, so a repeated system prompt pays in full on every call. A reasoning-tuned model naturally produces long outputs, which is the behaviour the price shape punishes hardest, and the usual escape route of caching the repeated block is closed.
What this usually looks like
- Output tokens per call here run well above your other open-weight models.
- A repeated system prompt goes out on every call with no cached rate to reduce it.
- Reasoning was enabled once for quality and its cost was never separated from the answer's.
- This model's bill is dominated by output while its call count looks unremarkable.
Free, no card, no account
Run the free Cost Leak Scan
It shows your most expensive conversation before you install anything.
Mistakes that cost the most
| Mistake | Why it hurts | Do instead |
|---|---|---|
| Treating this like the other open-weight models on cost shape. | Its five-to-one spread is closer to a frontier model than to the flat ratios beside it. | Price it on the output side and set a max token limit that reflects what you render. |
| Leaving reasoning on for tasks that don't need it. | Reasoning tokens bill as output at five times input and most of them never reach a user. | Enable it per task rather than globally, and measure the quality difference on each. |
| Planning around a cached rate that isn't there. | A forecast assuming a cached discount reads low by whatever share of your input repeats. | Price the repeated prefix at full input rate, then decide whether to shorten it instead. |
Run this check tonight
- Compare output tokens per call here against your other models on similar work.
- Separate reasoning output from delivered answer length. Only one of them reaches a user.
- Price your repeated prefix at full input rate, since no cached rate applies.
- Test the same task with reasoning off and score the difference honestly.
- Check whether a flatter-ratio model on the same platform would do this work.
What reasoning costs on a five-to-one spread
Illustrative example
12 million input tokens a month, of which 9 million is a repeated prefix, and 4 million output, on Qwen3.6 27B at real Groq rates. Volumes are modelled.
Output is already most of the bill before reasoning, and enabling it adds another 63%. With no cached rate to work on the input side, the only lever left is how much the model says.
Every number, with its confidence and source
| Figure | What it means | Confidence | Source |
|---|---|---|---|
| $0.60 / $3.00 per million | Qwen3.6 27B input and output rates on Groq | calculated | Price book row for groq/qwen/qwen3.6-27b, $0.0006 and $0.003 per 1k, effective 2026-07-02, re-verified against Groq's rendered pricing page 2026-08-01. |
| 5.00x | output rate as a multiple of input, the widest of any Groq model in this price book | calculated | $3.00 divided by $0.60, from the same price-book row. |
| $19.20 to $31.20 | modelled monthly cost with reasoning off and on, at identical input volume | estimated | Both endpoints from the teardown arithmetic at real Groq rates. A range because the token volumes are modelled. |
What a generic answer can’t know
A rate card can tell you output costs five times input. It can't tell you how much of your output is reasoning nobody reads, or which feature turned it on. That split lives in your own call records. Culpa keeps them on your infrastructure, where your prompts and responses stay, and counts the calls to run your plan.
Questions founders ask next
How much does Qwen3.6 27B cost on Groq?
$0.60 per million input tokens and $3.00 per million output, re-verified against Groq's pricing page on 2026-08-01. The price book records no cached rate for it and Groq's page lists none.
Why is this model's price shape awkward?
Because a five-to-one output spread and no cached rate work against each other. Long responses are expensive and the usual fix of caching the repeated input isn't available, so both of the standard levers are weakened at once.
Do reasoning tokens count as output?
On most providers, yes, and they bill at the output rate whether or not a user ever sees them. On a five-to-one spread that makes reasoning the single most expensive setting you can leave switched on by default.
On your infrastructure
Culpa runs on your infrastructure. Your prompts and responses never leave it. Culpa counts calls to run your plan, and it fails open, so if it ever breaks your app keeps running.
How Culpa works
Find the culprit. Not just the total.
Your dashboard shows what you spent. It stops short of who spent it. Culpa shows the conversation, the user and the feature behind it.
Your prompts stay local.
Culpa runs on your own infrastructure. What you send to a model reaches us at no point.
Every dollar has a name.
Follow any charge to the conversation, the user, the feature and the customer behind it.
See the bill before it lands.
Cost your next feature before you ship it. You get the likely bill and the worst case, at best, median, p90 and p99.
Three steps to your first answer.
Change one base URL.
Or drop in the Python or TypeScript library.
Find your most expensive conversation.
In the first session, not the first week.
Cost your next feature before you ship it.
Why the bill went up
Example dashboardCalls traced
418,209
across 3 projects
Spend this week
$378.41
+ $182 vs last week
Failed calls
312
74% retried, and you paid for all of them
+ $182 this week traced to one culprit
Spend over 14 days
Most expensive users
Next week forecast
Graded against reality. Accuracy shown as results land.
Free, no card, no account
Run the free Cost Leak Scan
It shows your most expensive conversation before you install anything.
Keep reading
Sources: Groq pricing. Last reviewed 2026-08-01, rates effective 2026-07-02. Plain text version.