Llama
Meta's open-weight LLM family — Llama 4 landed, Behemoth didn't
Overview
Llama is Meta's open-weight model line, now on Llama 4 (Scout/Maverick, April 2025) after Llama 2/3 seeded the open-source AI ecosystem. Behemoth was delayed repeatedly and never shipped; Meta pivoted its frontier bets to the closed-weight Muse Spark in April 2026.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 11 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
Use cases
ASEAN Perspective
Llama in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
Llama is still the backbone of the open-weight LLM ecosystem — Llama 2 and 3 built it, and Llama 4 (Scout and Maverick, shipped April 2025) kept it going, with a huge footprint on Hugging Face, Ollama, Together AI and Fireworks. But the story got messier in 2025-26: Meta got caught submitting a tuned, non-public 'chat' variant of Maverick to the LMArena leaderboard, and the publicly downloadable weights ranked roughly 32nd once tested honestly. Behemoth, the flagship trillion-plus-parameter model, hit training problems, slipped repeatedly, and was never released. In April 2026 Meta Superintelligence Labs shipped Muse Spark, a closed-weight, API-only model — a real pivot away from the open-weight strategy that made Llama famous. The licence also still isn't OSI-approved open source: a 700M-MAU commercial cap and EU restrictions apply. Good for self-hosting and fine-tuning; no longer a sure bet as Meta's frontier line.
What people say
April 2025 was Llama's high point and its credibility problem, in the same breath. Meta shipped Llama 4 Scout and Maverick that month, then got caught gaming the leaderboard: a special 'experimental' build of Maverick — tuned for longer, emoji-heavy answers that human raters tend to prefer — scored 1417 Elo on LMArena, while the actual public weights, tested afterward, landed around 32nd place on the same board. Independent testers also found Scout's real long-context performance nowhere near its marketed 10M-token window; one widely cited benchmark put it at 15.6% accuracy at 128k tokens versus Gemini 2.5 Pro's 90.6%.
Behemoth, the roughly 2-trillion-parameter flagship meant to anchor the family, never made it out the door. Meta blamed mid-training routing changes and chunked-attention blind spots for stalling its gains, pushed the release from summer to fall 2025, then quietly let it slip off the roadmap without a formal cancellation.
The bigger shift landed in April 2026: Meta Superintelligence Labs released Muse Spark, a closed-weight, API-only model, marking Meta's first real move away from the open-weight strategy Llama built its reputation on. Llama itself isn't dead — the model family still has enormous third-party infrastructure (Hugging Face, Ollama, Together AI, Fireworks) and remains a default choice for teams that want to self-host or fine-tune without per-token costs. But calling it 'the foundation of open-source AI' reads differently now that Meta's own frontier bets have moved behind closed doors, and the license was never OSI-recognized open source to begin with — a 700M-MAU commercial cap and EU access restrictions still apply.
Summary of public user & expert reviews, compiled by RECATOOLS.
Notable facts
- The Llama 1 model weights were leaked on 4chan within days of being shared with researchers — accidentally accelerating open-source AI adoption by months.
- Meta's Llama 3.1 405B was the first openly available model to match GPT-4 on standard benchmarks, proving that open-weight models could reach frontier performance.
- Llama models power over 25,000 fine-tuned variants on Hugging Face, making it the most forked AI model in history.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including Llama's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Llama unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Llama directly →
Spotted something out of date? Suggest an update →
Llama in the news
More in LLMs & Chat
