Skip to content
ReviewRamble

The best American open-source LLMs in 2026

The best American open-source LLMs right now are Gemma 4, Muse Glimmer and Nemotron 3.5 Super VL, out of 12 we have reviewed. Each is ranked by our rating, with its price, our verdict and who it suits.

By Andrew Miller · Updated Oct 26, 2026 · How we review

How we ranked: by our rating out of 5, from reviews built on each product's own pricing, documentation and policies plus independent benchmarks where they exist, each checked on the date shown in the review. We take no affiliate commission.

Also ranked: the best open-source LLMs

  1. 1

    Gemma 4

    Licence and data: Self-hosted, so data stays where you run it. Apache 2.0 allows commercial use and fine-tuning with no revenue cap or acceptable-use addendum.

    Running it: Weights are free. The 31B needs about 62 GB in BF16 (one 80 GB card) or 17.7 GB at 4-bit; 26B A4B is 14.4 GB, 12B 7.0 GB and E4B 5.2 GB at 4-bit. OpenRouter lists 31B at $0.20/$0.40 and 26B A4B at $0.08/$0.26 per 1M tokens.

    Price: Free weights under Apache 2.0. Hosted on Google Cloud and third parties; OpenRouter lists 31B at $0.20/$0.40 and 26B A4B at about $0.08/$0.26 per 1M tokens.

    Read our Gemma 4 review
    4.4/5
  2. 2

    Muse Glimmer

    Licence and data: Apache 2.0 permits commercial use with no revenue cap. Run locally, nothing leaves the machine; on OpenRouter the hosting provider's terms apply. Meta's training data includes information from its own products and services.

    Running it: Local: one 24GB card for the 17GB build, 32GB for the near-lossless build, 64GB for full precision. Hosted: $0.30/$1.20 per 1M tokens on DeepInfra, $0.35/$1.50 on Together, so 300M in and 60M out a month is about $162.

    Price: Free weights under Apache 2.0; no Meta API. Hosted on OpenRouter from $0.30 per 1M input and $1.20 per 1M output tokens (DeepInfra).

    Read our Muse Glimmer review
    4.1/5
  3. 3

    Nemotron 3.5 Super VL

    Licence and data: OpenMDW-1.1 permits commercial use. Self-hosting keeps data in-house; hosted providers set their own retention terms, and OpenRouter does not offer EU in-region routing for this model.

    Running it: On OpenRouter's $0.15/$0.50, 1M agent turns a month at 8K input and 1K output tokens comes to about $1,700, before its verbosity inflates output. Self-hosting at BF16 means four H100s per replica; NVFP4 fits a smaller Blackwell footprint.

    Price: Free weights under OpenMDW-1.1. Hosted: OpenRouter from $0.15 input / $0.50 output per 1M tokens, plus a free variant; NVIDIA NIM and other providers priced separately.

    Read our Nemotron 3.5 Super VL review
    4.1/5
  4. 4

    gpt-oss-120b

    Licence and data: Apache 2.0 with no usage restrictions in the licence. Self-hosted, prompts stay on your hardware; hosted, retention follows each provider's terms, and OpenRouter's EU-only routing needs a Business or Enterprise plan.

    Running it: 1B input and 250M output tokens a month cost about $300 at $0.15/$0.60 per million; Bedrock batch halves its rate. Self-hosting needs one 80GB GPU for inference and an H100 node to fine-tune.

    Price: Free weights under Apache 2.0; hosted prices vary by provider.

    Read our gpt-oss-120b review
    4.0/5
  5. 5

    Granite 4.2

    Licence and data: Apache 2.0, open for commercial and research use with no field-of-use limits; self-hosted prompts stay on your hardware.

    Running it: IBM's API charges $0.16 per million input and $0.65 per million output tokens, with a 75% discount on cache hits. Self-hosted, the 30B needs about 18GB at 4-bit or 31GB at 8-bit.

    Price: Free weights under Apache 2.0; watsonx and third-party hosting billed separately.

    Read our Granite 4.2 review
    4.0/5
  6. 6

    Muse Spark Open

    Licence and data: Muse Model License: free use, modification, distillation and redistribution; attribution required in each product; model-as-a-service providers above $20M annual revenue need a separate agreement. Meta claims no ownership of outputs. Self-hosted, nothing leaves your infrastructure.

    Running it: Self-hosting: NVFP4 (about 537GB) on one four-GPU Blackwell node, BF16 (about 1.65TB) on eight B300s, or two 512GB M5 Ultra Mac Studios. Hosted, a million support replies a month at 3,000 input and 500 output tokens is about $5,875 at OpenRouter rates.

    Price: Free weights under the Muse Model License (attribution required; $20M revenue threshold for model-as-a-service). Hosted on OpenRouter at $1.25 input, $0.15 cached, $4.25 output per 1M tokens.

    Read our Muse Spark Open review
    4.0/5
  7. 7

    Beam

    Licence and data: Apache 2.0, with Reflection's Usage Policy recommended alongside it. On Reflection's API, each response's reflection-data-retention header reports zero, standard or contributor retention; only contributor terms, which an organisation must accept, allow training on your requests.

    Running it: 1 billion input and 100 million output tokens a month cost about $520 on Together AI or Baseten and about $350 on Nebius. Self-hosting needs 273 GiB for NVFP4, 468.72 GiB for FP8 or 933.81 GiB for BF16 before any KV cache.

    Price: Free Apache 2.0 weights. Reflection API free in public beta with daily limits. Hosted: Together AI and Baseten $0.40 in / $0.04 cached / $1.20 out per 1M tokens; OpenRouter from $0.24 / $0.98.

    Read our Beam review
    3.8/5
  8. 8

    gpt-oss-20b

    Licence and data: Self-hosted, so prompts stay local. Apache 2.0 allows commercial use and fine-tuning; the usage policy only requires compliance with applicable law.

    Running it: Weights are free and fit a 16GB GPU. Hosted on OpenRouter from $0.018 input and $0.09 output per 1M tokens, so 1M short replies a month costs well under $100; Artificial Analysis puts it at about $0.01 per index task.

    Price: Free weights under Apache 2.0. Hosted from $0.018 input / $0.09 output per 1M tokens on OpenRouter; Artificial Analysis lists $0.07 / $0.25.

    Read our gpt-oss-20b review
    3.8/5
  9. 9

    Ornith-1.5-122B

    Licence and data: Self-hosted only, so data stays wherever you run it. There is no first-party API, and so no retention or training terms to review.

    Running it: Weights free. BF16 is 250 GB (a multi-GPU node), FP8 133 GB (two 80 GB cards), and NVFP4 (77.7 GB) or Q4_K_M (75.8 GB) the only builds near one 80 GB card; Macs get a 68.7 GB MLX 4-bit build. No first-party hosted price published; NanoGPT lists $0.20 input and $0.80 output per million tokens but marks the model unavailable.

    Price: Free open weights under MIT. You pay only for your own hardware or hosting.

    Read our Ornith-1.5-122B review
    3.6/5
  10. 10

    SmolLM3-3B

    Licence and data: Apache 2.0 with no acceptable-use addendum; commercial use and redistribution allowed. Self-hosted only, so data stays on your device or server. The training data mixture is public, which makes provenance checks possible.

    Running it: About 3.1B parameters, roughly 6GB in BF16, so a single consumer GPU, an Apple laptop or a 4-bit phone build. With no paid API listed, cost at volume is your own hardware.

    Price: Free weights under Apache 2.0, with the training data recipe and checkpoints published. No hosted API on Hugging Face's own Inference Providers.

    Read our SmolLM3-3B review
    3.5/5
  11. 11

    Phi-4-Reasoning-Vision-15B

    Licence and data: MIT licence; self-hosting keeps images and prompts local, and Foundry hosting follows Azure's terms.

    Running it: Self-hosting needs a GPU with well over 30GB, such as a 48GB A6000 or an 80GB A100; no Foundry price is published: neither the model card nor Azure's retail price list, which carries per-token meters for other Phi-4 models, lists one.

    Price: Free weights under MIT; Microsoft Foundry hosting billed separately.

    Read our Phi-4-Reasoning-Vision-15B review
    3.4/5
  12. 12

    Phi-4-multimodal

    Licence and data: MIT licence with no revenue cap or separate use policy. Self-hosted, audio and images stay on your hardware; Azure's terms apply only if you use Microsoft's hosted endpoint.

    Running it: On Azure AI Foundry, audio input costs $4 per 1M tokens against $0.08 for text and images, so speech dominates any bill; output is $0.32 per 1M, the same rates as at launch per Azure's retail price list. Self-hosting needs one recent data-centre GPU.

    Price: Free weights under MIT. Azure AI Foundry: $0.08 input / $0.32 output per 1M text or image tokens; audio input $4 per 1M.

    Read our Phi-4-multimodal review
    3.3/5

Sources

The main official source behind each ranked review; every review lists its full set.

More guides