Logo
Meta · Llama 4 Maverick · Apr 2025

Llama 4 Maverick — multimodal flagship of Meta’s open-weight series

  • Million-token context
  • Text and image together
  • Multilingual text generation
  • Downloadable weights to compare
Llama 4 Maverick

What is Llama 4 Maverick?

Llama 4 Maverick is Meta’s Llama 4 model from April 2025—built to understand text and images in one thread, with million-token context. Versus Llama 3.x, multimodal and long-context ability are fuller, and weights are downloadable under license for comparison. It leans general image-text work; for today’s long-task coding ceiling, compare Claude Opus 5, GPT-5.6 Sol, or Kimi K3. Pick Llama 4 Maverick in iMini Agent to use it.

Vendor
Meta
Released
Apr 2025
Context
1M tokens
Max output
16K tokens
Input modes
Text and image
Best for
Image-text understanding & open-weight comparison

What’s new vs Llama 3.x

Joint image-text reading, million-token context, and downloadable weights—the changes to check before you pick Maverick.

Text and image in one thread

Images and text enter the same understanding loop—ask with screenshots and charts together.

Million-token context

Long materials can stay in one thread—for documents and multi-turn chat.

Higher capacity in the same family

Versus Scout in the same wave, Maverick has higher capacity for heavier general image-text jobs.

Downloadable weights to compare

Obtain weights under the Llama 4 Community License for research and deploy evaluation; you can still run it on iMini online.

Official benchmarks

Architecture and long-context charts from Meta’s Llama 4 launch. Maverick is a native multimodal, 1M-context open-weight model.

Llama 4 Mixture of Experts diagram: routed and shared experts
Mixture of Experts. After attention, routed and shared experts run in parallel, then sum into the next layer—about Maverick’s multi-expert sparse activation, not a single score.
Llama 4 Maverick / Scout needle-in-a-haystack benches
Needle-in-a-haystack. Maverick text retrieval covers 1M nearly all green; Scout shows 10M text and up to ~20 hours of video retrieval. More direct than the nominal window for whether long context works.
Cumulative average negative log-likelihood on code sequences by position
Cumulative average NLL on code. As position stretches into the tens of millions, the curve keeps falling without rising back—prediction of later tokens does not collapse at extreme length.

Three common workflows

Llama 4 Maverick fits open-weight stacks and general multimodal jobs.

Image-text understanding and rewrite

Image Q&A, explainers, and multilingual text generation.

Long-document reading

Manuals and multi-file summaries. Note the tighter single-output limit vs today’s flagships.

Open-weight comparison

Evaluate whether the Llama stack meets delivery; for long-horizon Agent-specialized jobs, pick a current flagship instead.

Vs GPT-5.6 Sol and Claude Opus 5

All three are usable. Gaps are open weights, Agent specialization, and generation.

DimensionLlama 4 MaverickGPT-5.6 SolClaude Opus 5
Context1M tokens1M tokens1M tokens
ReasoningGeneral multimodalMax depth · ultraDeep thinking built in
Coding / AgentsGeneral capacity, not 2026 Agent-specializedHeaviest long-horizon coding todayEveryday long-horizon coding flagship
Long tool runsMore general chat and image-textFlagship depth + ultraStronger goal holding
Speed postureSelf-managed open-weight deployCompletion quality firstStability first
Prefer whenNeed the Llama weight stackHeaviest OpenAI depth-stack jobsClaude long-horizon workhorse

Using Llama 4 Maverick on iMini

Free quota, model identity, key specs, and how it differs from today’s flagships.

More models

More flagship models by vendor

GPT-5.6 SolOpenAI’s depth flagship—when the team is on the GPT stack and needs maximum depth.Claude Opus 5Anthropic’s Opus flagship—for long-horizon coding and multi-step Agents on the Claude stack.Gemini 3.1 ProGoogle’s Gemini Pro tier—for multimodal, long-context work on the Google stack.Grok 4.5xAI’s Grok flagship—when you need Grok reasoning or the xAI product stack.DeepSeek V4 ProDeepSeek’s coding and reasoning flagship—a strong open-weight pick for code and hard reasoning.Kimi K3Moonshot’s Kimi flagship—for CN/EN long context and China-market delivery.GLM 5.2Zhipu’s GLM flagship—for bilingual Agents and Zhipu-stack delivery.Qwen3.7 MaxQwen Max—when you need peak Qwen text reasoning.MiniMax M3MiniMax’s flagship—when delivery is tied to the MiniMax stack.Seed 2.0 LiteByteDance Seed’s lighter tier—for high-turnaround, shorter sessions.
All models

Use Llama 4 Maverick free on iMini Agent

Million-token context with text and image together—for image-text understanding and weight comparison.

Open Llama 4 Maverick

No Meta API key required.