Prototype · closed beta
unOpen.AI
In closed beta: Roxy, a private AI assistant with permanent memory running on infrastructure the user owns. Brand, site, product, backend and conversational design.
Model families
Meta models on
your own hardware.
Llama is the default answer to “we need to run this ourselves” — not because it always wins on capability, but because it has the deepest tooling ecosystem of any open-weight family, and on a self-hosted deployment the tooling is most of the work.
All Model Families →Introduction transcript. Qubi: When data can't leave your building, the model has to live there too. Someone qualified reads the licence first, and we try the smaller size before the larger. We compress it carefully, checking quality on your own examples as we go. Then the serving work: batching, health checks, upgrades, and a way back.
Current release
Meta today.
Meta ships these models as a range rather than one lead model, so this page is about the family: what it is good at, where it runs and how we evaluate it.
Last reviewed on 1 September 2026. We check the public model catalogues every week; when a new release appears, we review it against the provider's announcement and update this page.
LATEST MODELS
The newest Meta models.
What Meta has shipped most recently. Each has its own page with specifications, independent ratings at every reasoning setting and listed prices.
15 Oct 2026 · open weights
Muse Spark Open
- Intelligence
- 24.7
- Context
- 1.05M
- In / out per M
- $1.25 / $4.25
12 Oct 2026
Muse Wonder
- Intelligence
- 43.3
- Context
- 1.05M
- In / out per M
- $4.00 / $20.00
2 Sep 2026
Muse Spark 1.3
- Intelligence
- 36.4
- Context
- 1.05M
- In / out per M
- $1.25 / $4.25
9 Aug 2026 · open weights
Muse Glimmer 30B
- Intelligence
- —
- Context
- 131k
- In / out per M
- $0.300 / $1.20
EVERY MODEL
All 13 Meta models, compared.
Sort by any column. Select a model for its full page.
| Muse Spark Open open | 15 Oct 2026 | 24.7 | 140 | 1.05M | $1.25 | $4.25 |
|---|---|---|---|---|---|---|
| Muse Wonder | 12 Oct 2026 | 43.3 | 150 | 1.05M | $4.00 | $20.00 |
| Muse Spark 1.3 | 2 Sep 2026 | 36.4 | — | 1.05M | $1.25 | $4.25 |
| Muse Glimmer 30B open | 9 Aug 2026 | — | — | 131k | $0.300 | $1.20 |
| Muse Spark 1.2 | 5 Aug 2026 | — | — | 1.05M | $1.25 | $4.25 |
| Muse Spark 1.1 | 9 Jul 2026 | — | — | 1.05M | $1.25 | $4.25 |
| Llama 4 Maverick open | 5 Apr 2025 | — | 49 | 524k | $0.188 | $0.653 |
| Llama 4 Scout open | 5 Apr 2025 | — | 135 | 1.31M | $0.100 | $0.300 |
| Llama 3.3 70B Instruct open | 6 Dec 2024 | — | — | 131k | $0.100 | $0.320 |
| Llama 3.2 1B Instruct open | 25 Sep 2024 | — | — | 60k | $0.027 | $0.201 |
| Llama 3.2 3B Instruct open | 25 Sep 2024 | — | — | 131k | $0.050 | $0.330 |
| Llama 3.1 70B Instruct open | 23 Jul 2024 | — | — | 131k | $0.400 | $0.400 |
| Llama 3.1 8B Instruct open | 23 Jul 2024 | — | — | 131k | $0.050 | $0.080 |
Intelligence and coding: Artificial Analysis indices at the model's best reasoning setting, and its measured output speed. Context and prices: the OpenRouter catalogue list price in USD per million tokens. Released: the provider's date where Artificial Analysis records one, otherwise the date the catalogue first listed it. Refreshed weekly; not measured by QQuantum.ai. Select a heading to sort.
Specification
- Provider
- Meta, United States
- Weights
- Published and downloadable
- Licence
- Meta's community licence, not OSI open source. Commercial use is permitted with conditions, including an attribution requirement and a threshold on monthly active users above which separate terms apply. Read it; it is short, and it is not Apache-2.0. As of the version current at the time of writing; confirm against the provider's current terms. Not legal advice.
- Sizes
- A range from small models that run on a single accelerator to large ones needing a multi-GPU node
- Ecosystem
- The broadest of any open family — vLLM, Ollama, llama.cpp, quantisation formats, fine-tuning toolchains, cloud model catalogues
- Where we use it
- On-premise and private-cloud deployments, and as the baseline any other open model has to beat
Provider-published specifications as of 1 September 2026, not figures we measured. Model families move quickly — check the provider's own documentation before committing to a number.
Why the ecosystem is the argument
On a hosted API the model is the product. On a self-hosted deployment the model is one component of a system that also has to do batching, KV caching, quantisation, tensor parallelism, health checks, GPU scheduling, upgrades and rollback. Llama's advantage is that every one of those has a well-trodden path, several implementations and a large body of people who have hit the failure mode before you.
That matters more than a benchmark point. A model that scores marginally higher but has one immature serving implementation will cost you more in engineering time than the difference is worth, and it will cost you more again the first time you need to debug throughput at two in the morning. When we recommend against Llama it is usually because a specific capability requirement points elsewhere, not because the deployment story is better anywhere else.
The licence deserves ten minutes. It permits commercial use and it is not restrictive for the overwhelming majority of deployments, but it is a bespoke licence rather than a standard one, it carries an attribution obligation, and it has a scale threshold. If you are building a product on top of it, have someone read it who is qualified to. If a permissive standard licence is a hard requirement — and for some procurement processes it is — Mistral publishes Apache-2.0 options.
The practical levers
-
Pick the size from the task, not from the headline.
The mid sizes handle the overwhelming majority of production work, and the difference in hardware between a mid-size model and the largest is the difference between one node and several. Test the smaller one first; the result is frequently that it was enough. -
Quantisation is the cheapest capacity you will buy.
Running at eight or four bits cuts the memory requirement proportionally, and on many tasks the quality difference does not register on a real evaluation set. It has to be measured rather than assumed, and measuring it takes an afternoon. -
Serving configuration outweighs model choice.
Continuous batching, paged attention and prefix sharing decide how many concurrent users a node supports, and the gap between a naive loop and a properly configured vLLM deployment is a multiple rather than a percentage. See inference optimisation. -
Distil where the task is narrow.
Using a large model's outputs to train a small one converts capability you are renting into capability you own. Check the licence permits it for the specific model you deploy.
Where we would choose something else
Where a standard permissive licence is a procurement requirement. Llama's community licence permits commercial use but is bespoke, carries attribution obligations and has a scale threshold. Mistral publishes Apache-2.0 options and typically clears a standard open-source policy far faster.
Where reasoning-heavy work dominates and inference economics matter, DeepSeek's mixture-of-experts design frequently wins on throughput per unit of compute. Where the deployment has to span an edge device and a server with one tokeniser, Qwen's size ladder is the better fit.
And where nobody wants to operate inference infrastructure at all, a hosted API is cheaper once you price the people honestly. Self-hosting is a standing commitment, and we would rather say that in the first meeting than in the retrospective.
INDEPENDENT RATINGS
How Meta compares.
Third-party scores and listed prices, refreshed every week. Useful for building a shortlist; your own evaluation decides.
Composite index across their reasoning, knowledge, maths and coding evaluations · Higher is better · Linear scale from 0 to 66.0
- Gemini 4 Carbon
- 59.4
- Claude Opus 5.5
- 58.4
- GPT-6 Astra
- 56.9
- Muse Wonder Full
- 49.7
- Kimi K3.1 Pro
- 46.0
- Muse Wonder
- 43.3
- Muse Spark 1.3
- 36.4
- Muse Spark Open
- 24.7
This family's highest-rated models, highlighted, against the best-rated model of each other family we cover. One composite score cannot show cost, latency, data residency or how a model handles your task.
Source: Artificial Analysis. Checked 30 October 2026. Measured independently by Artificial Analysis, not by QQuantum.ai or the providers.
USD per million output tokens · The newest releases in the family · Linear scale from 0 to $22.00
- Muse Spark Open
- $4.25
- Muse Wonder
- $20.00
- Muse Spark 1.3
- $4.25
- Muse Glimmer 30B
- $1.20
- Muse Spark 1.2
- $4.25
Input is priced separately: Muse Spark Open $1.25, Muse Wonder $4.00, Muse Spark 1.3 $1.25, Muse Glimmer 30B $0.30, Muse Spark 1.2 $1.25 per million tokens. Catalogue list prices; the provider's own rates, caching and batch discounts can differ.
Source: OpenRouter model catalogue. Checked 30 October 2026. Listed by OpenRouter, not measured by QQuantum.ai.
Questions
Questions about Meta Llama & Muse.
Is Llama open source?
Not by the OSI definition, and the distinction is worth keeping straight because procurement sometimes cares. The weights are published and commercial use is permitted, which is what most people mean when they say open. But the licence carries conditions a standard open-source licence does not, including attribution and a user-count threshold. “Open weights” is the accurate term.
What does it cost to run?
Dominated by GPU capacity and the people who look after it, not by anything per-token. That flips the economics against hosted APIs only above a certain sustained volume — below it, you are paying for idle hardware. We work the crossover out explicitly rather than assuming it; see GPU sizing and cost.
Llama or DeepSeek for self-hosting?
DeepSeek tends to lead on reasoning-heavy work and has favourable inference economics from its mixture-of-experts design. Llama has the deeper tooling ecosystem and a US-domiciled provider, which some organisations' governance policies prefer. Both are credible; we benchmark both when the decision is open.
