Guides / llama 3.1 8b pricing
Llama 3.1 8B pricing, and the waste that hides behind it
Llama 3.1 8B pricing on Groq is $0.05 per million input tokens and $0.08 per million output tokens, the cheapest rate point in this price book. A single call rounds to a hundredth of a cent, which is precisely why nobody instruments it. Culpa, a local-first LLM cost, margin, and forecast ledger, counts every call including the ones too cheap to notice.
Why this happens
A model this cheap changes behaviour rather than budgets. Teams stop reviewing its call count, copy retry policies onto it from expensive models, and point it at every high-volume filtering job that comes along. None of that shows up per call, because per call there's nothing to see. It shows up as a monthly line nobody can account for, and the waste ratio inside it tends to be worse than on any model anyone watches.
What this usually looks like
- Nobody reviews this model's call count because the per-call cost looks like zero.
- Retry settings here were copied from a frontier model and never re-examined.
- It serves several high-volume jobs that were added without a cost review.
- Its monthly total surprises people the first time somebody adds it up.
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 a cheap model as exempt from instrumentation. | Cheap times enormous volume is a real number, and it's the least examined one you have. | Meter it like anything else. The cost of counting is the same regardless of the rate. |
| Copying a five-attempt retry policy onto a high-volume cheap model. | The multiplier applies to a call count several orders of magnitude larger than the model it came from. | Set retries from the failure mode and the volume, not from what the expensive model uses. |
| Adding jobs here without a review because the rate is negligible. | Every job added at negligible cost sums to a line item nobody owns. | Give the model an owner and a monthly budget, the same as a costly one. |
Run this check tonight
- Total this model's calls for last month. The count matters more than the rate.
- Compute cost per call to five decimal places, then multiply by the count. Read the total out loud.
- Find your retry setting here and multiply your worst case by it.
- List every job pointed at this model and check each had a cost review.
- Compare attempts against successful results. On a cheap model that ratio is rarely watched.
A rate that rounds to nothing, at twenty million calls
Illustrative example
A filtering job on Llama 3.1 8B Instant at real Groq rates of $0.00005 per 1k input and $0.00008 output. Each call sends 2,000 input tokens and returns 400. Call volume and the failure rate are modelled.
One call costs $0.000132 and the month costs $2,640, of which $317 is retries. The rate is what makes it invisible and the volume is what makes it matter.
Every number, with its confidence and source
| Figure | What it means | Confidence | Source |
|---|---|---|---|
| $0.05 / $0.08 per million | Llama 3.1 8B Instant input and output rates on Groq, the cheapest point in this price book | calculated | Price book row for groq/llama-3.1-8b-instant, $0.00005 and $0.00008 per 1k, effective 2026-07-02, re-verified against Groq's rendered pricing page 2026-08-01. |
| $0.000132 | modelled cost of a single 2,000-token call returning 400 tokens | calculated | (2 x $0.00005) + (0.4 x $0.00008) per 1k, using real Groq rates. Token counts are modelled. |
| $316.80 | modelled monthly retry waste at a 3% failure rate and a five-attempt ceiling | calculated | 2,400,000 wasted attempts x $0.000132, from the teardown arithmetic at real Groq rates. The failure rate and call volume are modelled. |
What a generic answer can’t know
A provider bills one total for this model and has no idea which of your jobs produced it, or how many of the calls were attempts rather than results. Both facts live in your application. Culpa records them on your infrastructure, keeps your prompts and responses there, and counts the calls to run your plan.
Questions founders ask next
How much does Llama 3.1 8B cost on Groq?
$0.05 per million input tokens and $0.08 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.
Is a model this cheap worth metering?
Yes, because the volume is what makes it expensive rather than the rate. A model at a twentieth of a cent per million tokens still produces a four-figure monthly line when it serves tens of millions of calls, and it's the line nobody reviews.
What usually goes wrong on a very cheap model?
Retry policy, most often. A five-attempt ceiling copied from a frontier model multiplies against a call count several orders of magnitude larger, so the waste ratio here tends to be worse than on any model anyone watches.
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.