Skip to content
Best AI Ever

Meta · AI model

Llama 4 Scout

Open-weight natively multimodal mixture-of-experts model (17B active / 109B total parameters, 16 experts) released April 2025 under the Llama 4 Community License. Text and image input, text and code output, 10M-token context.

Last verified Available 4 sources

Key specs

Provider
Meta
Family
Llama 4
Availability
Available
Release date
2025-04-05
License
Open weights (Llama 4 Community License)
Context window
10,000,000 tokens
Knowledge cutoff
August 2024
API access
Unknown
Consumer app
Unknown

Capabilities

Inputs

  • Text input Yes
  • Image input Yes Tested for image understanding up to 5 input images; image understanding is English-only (Llama 4 model card on dev.meta.ai)
  • Audio input No Input modalities listed as multilingual text and image
  • Video input No Input modalities listed as multilingual text and image
  • File input Unknown

Outputs

  • Text output Yes
  • Image output No Output modalities listed as multilingual text and code
  • Audio output Unknown

Features

  • Reasoning Unknown
  • Tool calling Yes Zero-shot function calling in Python-style and JSON formats; tool output uses the ipython role
  • Web search Unknown

"Unknown" means we have not verified this capability yet; it does not mean the model lacks it.

Pricing

No current prices listed.

Benchmarks

BenchmarkSettingScoreReported byMeasuredVerifiedSource
GDPval-AA v2.1 Elo, higher is better v2.1; no effort setting; Artificial Analysis Stirrup harness -164 Elo Third party 2026-10-25 2026-10-25 GDPval-AA v2.1 Leaderboard (Artificial Analysis)
Text Arena Arena score, higher is better Overall, style control on (default) 1,322 Third party 2026-10-23 2026-10-25 Text Arena leaderboard (Arena)
GDPval-AA v2.1 caveats:

Elo is a relative rating: values change as models are added and the scale is re-anchored, so always store the date a rating was read. For example, Anthropic's launch page gives 1846 for Claude Opus 5.5, while the leaderboard on 2026-10-12 shows 1866 (Max, Default Fallback). Earlier versions (Google cites "GDPVal-AA v2") are not comparable. Judging is done by an LLM, not by human experts. Anthropic models are listed with "Default Fallback", meaning refused steps are routed to a fallback model. AA-Briefcase and GDP.pdf are separate evaluations.

Text Arena caveats:

Scores measure voter preference, which can be influenced by answer style and length. The leaderboard offers a style-control adjustment, so record which view was used. Scores shift with every update, so a value is only meaningful together with its leaderboard date. Entries marked "Preliminary" or with few votes have wide confidence intervals. The same model can appear several times with different reasoning-effort suffixes (for example -high or -max). Category views (Coding, Math, Creative Writing and others) are different rankings from Overall.

One row per result, without rank numbers: each result was measured under its own setting (effort level, harness, tools, test version), as stated by the source, and scores compare only within the same setting. Self-reported results come from the provider and have not been reproduced independently.

Rankings

Derived from our published ranking snapshots; each ranking explains its criteria and weights.

Compare Llama 4 Scout

Guides

About Llama 4 Scout

Llama 4 Scout is a natively multimodal mixture-of-experts model with 17B active and 109B total parameters across 16 experts, released on 5 April 2025. It accepts multilingual text and image input and outputs text and code, with a 10M-token context window. Meta states it fits on a single H100 GPU with on-the-fly int4 quantization. Its knowledge cutoff is August 2024, and the weights are distributed under the Llama 4 Community License.

Strengths and weaknesses

Strengths

  • 10M-token context window (Llama 4 Maverick: 1M; Muse Spark: 1,048,576)
  • Open weights under the Llama 4 Community License; fewer total parameters than Maverick (109B vs 400B), fitting a single H100 GPU with int4 quantization
  • Native text and image input, with 12 officially supported languages
  • Documented zero-shot function calling (Python-style and JSON formats)

Weaknesses

  • Knowledge cutoff of August 2024
  • Image understanding is English-only and tested up to 5 input images
  • No Meta first-party API price listed; Hugging Face download requires accepting gated access conditions

Sources

  1. meta-llama/Llama-4-Scout-17B-16E-Instruct model card
  2. Llama 4 model card and prompt formats
  3. GDPval-AA v2.1 Leaderboard (Artificial Analysis)
  4. Text Arena leaderboard (Arena)

Last verified Page updated 2026-10-26