LaunchpadHQ
All model families

Meta

Llama (Meta) models

Version comparison

Last verified Oct 14, 2026. Figures that could not be confirmed on an official vendor page are marked unverified — we do not guess.

Full Llama (Meta) comparison

Which one should I use?

  • There is no Platform.slug in this repo’s tutorials or seed scripts for Meta’s Llama product (llama-cpp is a separate local engine). This family is documented for /models only.
  • Llama 4 Scout: long context (10M tokens on the model card) and can run on a single H100 when quantized.
  • Llama 4 Maverick: larger expert count (400B total / 17B active) and a 1M context window.
  • Meta does not list a first-party per-1M-token API price on the pages checked — you download weights or pay a host. API prices are unverified here.
  • Llama 4 is still Meta's newest Llama-branded release, but llama.com now leads with Meta's newer Muse models (checked 2026-10-24). See the Muse family for Muse Spark Open and Muse Glimmer 30B.

Llama 4 Maverick

Current
Context
1M
Max output
Unverifiedunverified
API input / 1M
Unverifiedunverified
API output / 1M
Unverifiedunverified

Best for: Teams that can host Llama themselves and want the stronger Llama 4.

Weights are downloadable from Meta and partners. Which Meta AI app experiences use Maverick was not confirmed on the pages checked.

Source: Llama 4 model card

Llama 4 Scout

Current
Context
10M
Max output
Unverifiedunverified
API input / 1M
Unverifiedunverified
API output / 1M
Unverifiedunverified

Best for: Very long documents on hardware you control, when you do not need Maverick.

Weights are downloadable. Meta AI app mapping was not confirmed on the pages checked.

Source: Llama 4 model card

Llama 4 Maverick

Teams that can host Llama themselves and want the stronger Llama 4.

  • Natively multimodal (text plus up to five images in, text out).
  • 1M token context on the official model card.
  • 17B active parameters from a 400B MoE (128 experts).

Caveats

  • Not a one-click chat app in this catalog — no matching platform slug.
  • API token prices are unverified (hosts set their own).
  • Max output unverified.

No first-party per-token API price on llama.com / the model card.

Llama 4 Scout

Very long documents on hardware you control, when you do not need Maverick.

  • 10M token context window on the official model card.
  • Fits on a single H100 at INT4, per Meta’s card.
  • Same 17B active-parameter MoE design, with 16 experts / 109B total.

Caveats

  • 10M context is an evaluated maximum; real hosts may expose a shorter window.
  • API token prices unverified.

No first-party per-token API price on llama.com / the model card.