Meta (Groq) · Llama 3.3 · Mid
Llama 3.3 70B pricing and limits
Meta's open-weight 70B model served on Groq's LPU hardware. Very fast output and output priced close to input; the same model is available from several hosts at different rates.
| Line | Tokens | Rate / 1M | Cost |
|---|---|---|---|
| Input, uncached | 6,000 | $0.59 | $0.00354 |
| Output | 700 | $0.79 | $0.00055 |
- Input / 1M
- $0.59
- Output / 1M
- $0.79
- Cached / 1M
- n/a
- Cache write / 1M
- n/a
- Blended 3:1
- $0.64
- Context
- 128K
What drives Llama 3.3 70B cost
Because Llama 3.3 is open weight, the price depends on the host. The index lists Groq and Together rates for the same model; the difference reflects hardware, speed and service terms rather than model quality.
There is no cache-read discount in the Groq list price for this model, so prompts are billed in full. High output speed means agents complete more steps per minute, which helps users and also makes loops run faster.
Cost of a typical call
Take a chat turn with 6,000 input tokens and 700 output tokens. At Llama 3.3 70B list prices ($0.59 input, $0.79 output per 1M) one call costs $0.00409, and 600,000 calls a month come to $2,456. The list price has no discounted cache-read rate, so every input token is billed at the full rate on every call.
Blended at a 3:1 input to output ratio, Llama 3.3 70B costs $0.64 per 1M tokens. Routing the same call to the fallback in the policy below, Llama 3.1 8B Instant, costs $0.00036, which is what a tenant pays you in COGS once they cross the hard cap and PALM SOUQ downgrades instead of blocking.
Cost comparison
Llama 3.3 70B against three alternatives
The same chat turn priced on Llama 3.3 70B Turbo, Llama 4 Maverick, GPT-5 mini. Move the sliders to match your traffic; models without a cache-read rate bill every input token in full.
Benchmark quality per task before switching. A cheaper model that needs more steps or longer outputs can cost more per finished task than its rate suggests, which is why PALM SOUQ reports cost per feature, not only per token.
Llama 3.3 70B this guide
$0.00409 per call
$2,047 / mo
baseline
Llama 3.3 70B Turbo
$0.00590 per call
$2,948 / mo
+44% vs Llama 3.3 70B
Llama 4 Maverick
$0.00162 per call
$810 / mo
-60% vs Llama 3.3 70B
GPT-5 mini
$0.00290 per call
$1,450 / mo
-29% vs Llama 3.3 70B
List prices from the PALM SOUQ LLM Price Index, computed per call as uncached input + cache reads + output. Reasoning tokens count as output on reasoning models.
Rate limits by usage tier
Groq sets limits per organization and per model as requests per minute and per day and tokens per minute and per day, with higher limits on paid tiers. Daily token limits can be the binding constraint for batch jobs.
Exact limits vary by account tier and change over time, so we do not reprint them here. Check the limits page for your own account in the Meta (Groq) documentation at console.groq.com/docs. PALM SOUQ enforces your own per-tenant limits in front of the provider's, so a single customer cannot consume the shared allowance that every other tenant depends on.
Circuit-breaker best practices
Token velocity matters most on fast inference: a loop on Groq can issue many more calls per second than on slower hosts, so set the velocity window tight.
Keep a cross-host fallback (for example Together) in the fallback chain so a Groq rate limit becomes a reroute instead of an outage.
When a breaker trips, the gateway answers in under 50 ms with HTTP 429 and X-PalmSouq-Reason: circuit_breaker_triggered, before the request reaches Meta (Groq). Your agent framework sees a normal rate-limit response and stops, and the audit log records which heuristic fired and for which tenant.
# palmsouq.policy.yaml
policy: llama-3-3-70b-groq-pro
match:
model: llama-3.3-70b-groq
tier: pro # from X-PalmSouq-Tier
budgets:
monthly_usd: 100 # per tenant
soft_cap_pct: 80 # webhook to your app
on_hard_cap: fallback # or: block (HTTP 429)
fallback:
- llama-3.1-8b-groq
breakers:
tool_thrashing: { identical_calls: 3, window_steps: 8 }
token_velocity: { max_tokens_per_s: 50000, window_s: 10 }
semantic_echo: { repeats: 3 }
max_steps: 15- budgetsPer-tenant monthly dollar cap with an 80% soft-cap webhook, keyed by X-PalmSouq-Tenant.
- fallbackAt the hard cap, requests reroute to Llama 3.1 8B Instant ($0.058 blended per 1M) instead of failing.
- breakersAll four loop heuristics on, evaluated in the request path with under 5 ms added latency.
Where it fits
Latency-sensitive chat and voice applications, and teams that want open-weight models with the option to switch hosts without changing prompts.
Provider documentation: console.groq.com/docs
Llama 3.3 70B pricing questions
How much does Llama 3.3 70B cost per million tokens?
At list price Llama 3.3 70B costs $0.59 per 1M input tokens and $0.79 per 1M output tokens. Prices were last reviewed in the PALM SOUQ LLM Price Index and are for the standard tier and shortest context band.
Does Llama 3.3 70B support prompt caching?
The Meta (Groq) list price for Llama 3.3 70B has no discounted cache-read rate, so PALM SOUQ meters every input token at $0.59 per 1M. Shortening repeated context and caching full responses in your own application are the main savings levers.
What are the Llama 3.3 70B rate limits?
Groq sets limits per organization and per model as requests per minute and per day and tokens per minute and per day, with higher limits on paid tiers. Daily token limits can be the binding constraint for batch jobs. For your account's exact numbers, see the Meta (Groq) documentation at console.groq.com/docs.
How do I stop runaway Llama 3.3 70B agent loops?
Token velocity matters most on fast inference: a loop on Groq can issue many more calls per second than on slower hosts, so set the velocity window tight. The sample policy on this page enables all four PALM SOUQ heuristics with a max step cap of 15 and falls back to Llama 3.1 8B Instant at the hard cap.
Is there a batch discount for Llama 3.3 70B?
Discounted asynchronous processing through the Groq Batch API. Batch traffic suits evaluations, backfills and nightly enrichment jobs that do not need a real-time answer.
Govern Llama 3.3 70B spend per customer.
Point your Meta (Groq) client at gateway.palmsouqltd.com, pass X-PalmSouq-Tenant, and every Llama 3.3 70B token is attributed, capped and protected by circuit breakers.
14-day free production trial · 100k requests included · no card required