Llama API Pricing (Meta)
We track 6 Llama models. The cheapest is Llama 3.1 8B at $0.020/MTok input and $0.040/MTok output, and the most expensive is Muse Spark 1.3 at $1.25/MTok input and $4.25/MTok output.
Prices checked 13 Nov 2026 against Meta’s pricing page ↗
| Model | Input | Output | Cached in |
|---|---|---|---|
| Llama 3.1 8BBudget | $0.020 | $0.040 | – |
| Llama 3.3 70BMid-tier | $0.10 | $0.32 | – |
Per 1M tokens · 6 current models · Largest context 10M tokens
Model prices
| Model | Tier | Input | Output | Over size | Cached in | Batch in / out | Context | Details |
|---|---|---|---|---|---|---|---|---|
| Muse Spark 1.3 | Mid-tier | $1.25 | $4.25 | — | $0.15 | — | 1M | Details for Muse Spark 1.3 |
Latest Muse Spark (Sep 2, 2026) on the Meta Model API, Meta's first paid first-party API. Same $1.25/$4.25 as 1.1 (Jul 9) and 1.2 (Aug 5), which stay available. Cached input $0.15. Multimodal agentic model with computer use. A Contributor tier bills $0.10/$0.20 (cached $0.002) in exchange for Meta training on your prompts and outputs, capped at 150 requests a minute. Still a public preview, but no longer US-only: access widened beyond the US with 1.2 on Aug 5. | ||||||||
| Muse Ember | Budget | $0.10 | $0.50 | over 200K: $0.30 / $1.50 | $0.0040 | — | 1M | Details for Muse Ember |
Meta's efficient execution model on the Meta Model API, built for high-volume sub-agent tasks, tool calls and routing. Listed here since Nov 10, 2026; Meta's pages give no launch date. Rates shown are for requests up to 200K input tokens: a request over 200K, cached tokens included, bills the whole request at $0.30/$1.50. Cached input $0.004 ($0.012 above 200K). Max output 524,288 tokens. Standard mode only: no Contributor tier and no batch discount. | ||||||||
| Llama 4 Scout | Budget | $0.10 | $0.30 | — | — | — | 10M | Details for Llama 4 Scout |
10M native context window, but DeepInfra, whose rate is shown, serves 320K. Hosts vary in both price and served context. | ||||||||
| Llama 4 Maverick | Mid-tier | $0.27 | $0.85 | — | — | — | 1M | Details for Llama 4 Maverick |
Rate shown is Novita's, with the full 1M context: DeepInfra, our usual reference host, retired Maverick on Oct 1, 2026. Varies by host. | ||||||||
| Llama 3.3 70B | Mid-tier | $0.10 | $0.32 | — | — | — | 128K | Details for Llama 3.3 70B |
Commodity host rates $0.10/$0.32 (DeepInfra). Varies widely by host. | ||||||||
| Llama 3.1 8B | Budget | $0.020 | $0.040 | — | — | — | 128K | Details for Llama 3.1 8B |
Llama 3.1 8B on hosted providers ($0.02/$0.04 on DeepInfra) — 128K context, the cheapest rate tracked. Supersedes the original Llama 3 8B (8K). | ||||||||
No models match those filters
What would Llama cost you?
Preset to Meta’s current models. Adjust to your usage.
- 1Llama 3.1 8B $0.0000 per request$2.30
- 2Llama 4 Scout $0.0003 per request$13.50
- 3Llama 3.3 70B $0.0003 per request$13.90
- 4Muse Ember $0.0004 per request$17.50
- 5Llama 4 Maverick $0.0007 per request$37.25
- 6Muse Spark 1.3 $0.0036 per request$178.75
50,000 requests × (1,500 in + 400 out). List prices; excludes tax and cache-write fees.
What Llama costs for real jobs
Three jobs priced for Meta’s budget, mid-tier and flagship models.
Support replies
1,000 replies a day for 30 days. 30,000 requests, 1,500 input and 300 output tokens each.
| Model | List price | Batch |
|---|---|---|
| Llama 3.1 8BBudget | $1.26 | – |
| Llama 3.3 70BMid-tier | $7.38 | – |
Coding agent session
One session of 40 requests that mostly re-read the same code. 40 requests, 60,000 input and 1,500 output tokens each; where a cached price exists, 80% of the input is billed at it.
| Model | List price | Batch | Cached (80% of input) |
|---|---|---|---|
| Llama 3.1 8BBudget | $0.050 | – | This model has no cached-input price in our data, so only the list total is shown. |
| Llama 3.3 70BMid-tier | $0.26 | – | This model has no cached-input price in our data, so only the list total is shown. |
Long document review
50 documents a month. 50 requests, 120,000 input and 2,000 output tokens each.
| Model | List price | Batch |
|---|---|---|
| Llama 3.1 8BBudget | $0.12 | – |
| Llama 3.3 70BMid-tier | $0.63 | – |
Totals are for all the requests in each example, in US dollars, from the same cost function as the calculator above.
Is there a free API tier?
Meta’s pricing page lists no free API tier.
Checked 11 Nov 2026 on Meta’s pricing page ↗.
Latest launches and price changes
- 10 Nov 2026 · launch · Muse Ember
Llama pricing FAQ
What is the cheapest Llama model?
Llama 3.1 8B is the cheapest current Llama model, at $0.020 per million input tokens and $0.040 per million output tokens.
Which Llama model should I use?
By tier, the models we list first are Llama 3.1 8B (budget) and Llama 3.3 70B (mid-tier): the cheapest budget and mid-tier models we track that are open to new users. To spend less, pick the tier that fits the job and its cheapest model: For everyday work, mid-tier models: Llama 3.3 70B, Llama 4 Maverick and Muse Spark 1.3 (from $0.10 / $0.32 for Llama 3.3 70B). For high-volume, simple jobs, budget models: Llama 3.1 8B, Llama 4 Scout and Muse Ember (from $0.020 / $0.040 for Llama 3.1 8B). Prices are input / output per million tokens. Tiers group models by capability, not price, so price your own workload in the calculator on this page.
Does Llama offer batch or caching discounts?
We list no batch discount for Meta's current models, so the list price is what you pay. Prompt caching: reading cached input costs 4% to 12% of the input price on 2 of 6 current models; Muse Ember's $0.10 input falls to $0.0040 per million tokens when read from the cache.
What context window do Llama models have?
The largest is 10M tokens on Llama 4 Scout, then 1M tokens on Muse Ember, Llama 4 Maverick and Muse Spark 1.3 and 128K tokens on Llama 3.1 8B and Llama 3.3 70B. A request's prompt and its output together must fit in the window.
What does the Llama API cost per person each month?
It depends on how much each person uses it. At list prices, an occasional user (3 short requests a working day) costs about $0.0015 a month on Llama 3.1 8B, and a daily power user (40 longer requests a day) about $0.92 on Llama 3.3 70B. The cost-per-seat calculator works it out for a whole team.
How often do Llama API prices change?
We check Meta's pricing page every day, and last re-verified every price on 13 Nov 2026. Since Feb 2026 we have recorded 4 price cuts and 3 price rises on Llama models, the latest on 1 Oct 2026, plus 5 launches. The price history chart on this page shows how each tier has moved.
Price history
Input price of the cheapest Llama model in each tier, month by month, among the models on today’s price list.
View as a table
| Month | Mid-tier | Budget |
|---|---|---|
| Feb 2026 | $0.28 | $0.10 |
| Mar 2026 | $0.28 | $0.10 |
| Apr 2026 | $0.28 | $0.10 |
| May 2026 | $0.15 | $0.080 |
| Jun 2026 | $0.15 | $0.080 |
| Jul 2026 | $0.15 | $0.030 |
| Aug 2026 | $0.10 | $0.020 |
| Sep 2026 | $0.10 | $0.020 |
| Oct 2026 | $0.10 | $0.020 |
| Nov 2026 | $0.10 | $0.020 |
Launches and price changes
- Llama 3.3 70B: input -19%, output -19%▼ Price cut
- Llama 3.3 70B: reverted to $0.88/$0.88 on Together.ai▲ Price rise
- Meta releases Llama 4 Scout (10M context) and finalises MaverickLaunch
- Llama 4 Maverick hosted rates fall to ~$0.15/$0.60 on DeepInfra/OpenRouter▼ Price cut
- Llama 3.3 70B: commodity host rates fall to ~$0.23/$0.40 (DeepInfra); $0.88 is now Together-only▼ Price cut
- Meta budget slot updated to Llama 3.1 8B — 128K context, ~$0.03/$0.06 on hosted providers (supersedes Llama 3 8B / 8K)Launch
- Meta opens its first paid first-party API — Muse Spark 1.1 at $1.25/$4.25 (1M context), US developers only in public previewLaunch
- Meta hosted rates re-baselined against DeepInfra: Llama 3.3 70B falls to $0.10/$0.32 (-57% input) and Llama 3.1 8B to $0.02/$0.04, the cheapest rate tracked▼ Price cut
- Meta hosted rates rise for the Llama 4 pair: Maverick to $0.20/$0.80 (+33% input) and Scout to $0.10/$0.30 (+25% input)▲ Price rise
- Meta releases Muse Spark 1.3 at an unchanged $1.25/$4.25; the tracked Muse Spark row moves from 1.1 to 1.3. Muse Spark 1.2 shipped Aug 5 at the same price and widened access beyond the USLaunch
- DeepInfra retires Llama 4 Maverick, so the tracked rate moves to Novita at $0.27/$0.85 (from $0.20/$0.80), keeping the full 1M context▲ Price rise
- Meta now caps Muse Spark Contributor mode at 150 requests a minute, up from 100Tier change
- Muse Ember: Meta now offers Muse Ember, its budget model on the Meta Model API, at $0.10/$0.50 for requests up to 200K input tokens and $0.30/$1.50 above (1M context)Launch