Guides / llama 3.3 70b pricing
Llama 3.3 70B pricing, and why the usual advice fails here
Llama 3.3 70B pricing on Groq is $0.59 per million input tokens and $0.79 per million output tokens, a ratio of 1.34 and the flattest in this price book. Shortening answers, the standard cost fix, returns almost nothing at that ratio. Culpa, a local-first LLM cost, margin, and forecast ledger, prices input and output separately so the advice gets tested rather than assumed.
Why this happens
Nearly every piece of LLM cost advice assumes output costs several times input, because on hosted frontier models it does. Trim the response, cap max tokens, stop the model narrating. All of it derives from a five to eight times spread. Here the spread is 1.34, so an output token costs about a third more than an input token rather than eight times more. The optimisation that saves a third of a Gemini bill saves a rounding error of this one, and the levers that do work are call count and input size.
What this usually looks like
- An output-trimming project shipped and the bill barely moved.
- Cost advice gets copied between models with very different ratios and nobody re-derives it.
- Input volume has never been examined because attention went to response length.
- Nobody can state this model's output-to-input ratio without going to look 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 |
|---|---|---|
| Applying frontier-model cost advice to a flat-ratio model. | The advice was derived from an eight-to-one spread and this model runs at 1.34 to one. | Compute the ratio first, then decide which lever is worth pulling. |
| Spending engineering time capping output here. | Halving output cuts roughly a seventh of the bill, which rarely repays the work. | Attack input size and call count. On a flat ratio those are almost the whole bill. |
| Assuming an open-weight model is too cheap to instrument. | Cheap per call plus high volume is exactly the shape that produces an unexplained line item. | Meter it like any other model. The rate is low and the volume usually isn't. |
Run this check tonight
- Divide the output rate by the input rate. Anything near one means output trimming is the wrong lever.
- Split last month's tokens into input and output and price each separately.
- Model what halving your output would actually save here, before committing anyone to the work.
- Do the same calculation on your highest-spread model. The contrast is the point.
- Check your input volume trend, which is what actually moves a flat-ratio bill.
The same optimisation, on two different ratios
Illustrative example
A workload of 20 million input and 6 million output tokens in a month, priced on Llama 3.3 70B at real Groq rates and on Gemini 2.5 Flash at $0.30 and $2.50 per million for contrast. Volumes are modelled.
The same optimisation is worth two and a half times more on the higher-spread model. Advice that travels between models without the ratio attached is advice about somebody else's bill.
Every number, with its confidence and source
| Figure | What it means | Confidence | Source |
|---|---|---|---|
| $0.59 / $0.79 per million | Llama 3.3 70B Versatile input and output rates on Groq | calculated | Price book row for groq/llama-3.3-70b-versatile, $0.00059 and $0.00079 per 1k, effective 2026-07-02, re-verified against Groq's rendered pricing page 2026-08-01. |
| 1.34x | output rate as a multiple of input, the flattest ratio in this price book | calculated | $0.79 divided by $0.59, from the same price-book row. |
| 14% to 36% | modelled saving from halving output, on this model against Gemini 2.5 Flash | estimated | Both endpoints from the teardown arithmetic at real rates for each model. A range because the token volumes are modelled. |
What a generic answer can’t know
A rate card gives you two numbers. Which lever is worth pulling depends on your own split between input and output tokens, and on which features generate which. Culpa measures that split per feature 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 3.3 70B cost on Groq?
$0.59 per million input tokens and $0.79 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 does trimming output save so little here?
Because output costs only about a third more than input on this model. On a frontier model where output bills eight times input, the same trim removes a much larger share of the bill. The ratio decides the value of the fix.
What should I optimise on a flat-ratio model instead?
Input size and call count. With output priced near input, a shorter system prompt or a removed redundant call moves the total about as much as the equivalent output reduction, and both are usually easier to ship.
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.