Built with Llama
Open Token offers products and services that use Meta Llama models.
Built with Meta Llama 3
1. How model terms apply
Use of a model is subject to Open Token's Terms and Acceptable Use Policy, plus the license, use policy, and provider terms identified here. Customer must review the current linked terms before using a model and must pass through restrictions to its own users when those terms require it. If a model term is stricter than Open Token's general terms, the stricter model term controls for that model.
Links point to documents maintained by the model owner or Amazon Web Services. Those documents may change independently of this page. Names, marks, and model copyrights belong to their respective owners. Listing a model does not imply endorsement or sponsorship.
2. Amazon Bedrock baseline
Models are served through Amazon Bedrock. Every request is subject to the AWS Service Terms, including Section 50, the AWS Acceptable Use Policy, the AWS Responsible AI Policy, and the applicable Bedrock third-party model terms.
- Do not use Outputs to train or develop a substantially similar model where AWS terms prohibit it.
- Do not extract model weights or training data.
- Do not remove provenance information, safety metadata, or watermarks.
- Customer must provide required privacy notices and obtain rights and consents for its Inputs.
3. Models available on Open Token platform keys
Amazon Nova
Nova Micro, Lite, Pro, Premier, and Nova 2 Lite are subject to the AWS baseline above.
Meta Llama 3, 3.1, 3.3, and 4
Meta community licenses and acceptable-use policies: Llama 3, Llama 3.1, Llama 3.3, and Llama 4
Customer must comply with Meta's applicable Llama Acceptable Use Policy and preserve required attribution. Organizations above the applicable 700-million-monthly-active-user threshold must obtain Meta's separate permission.
Due to the Llama 4 multimodal license restriction, companies headquartered in the European Union cannot use Open Token's provider account for Llama 4 Scout or Maverick. Those models remain available only through the company's own Amazon Bedrock credentials, subject to its direct rights.
Google Gemma 3 and Gemma 4
Gemma Terms of Use, Gemma Prohibited Use Policy, and Gemma 4 Apache 2.0
Gemma 3 4B, 12B, and 27B are distributed as a hosted API under the Gemma Terms. The Gemma Prohibited Use Policy is binding on Customer and its users. Customer must give downstream users notice of these terms. Gemma 4 models are offered under Apache 2.0 and require preservation of applicable notices.
Mistral open models
Mistral on Bedrock terms and applicable Apache 2.0 or model license
Mistral 7B, Mixtral 8x7B, Mistral Large 3, Ministral 3, Magistral Small, and Voxtral models preserve their applicable Apache 2.0 notices. Devstral 2 uses a modified MIT license; Open Token must stop offering it on platform keys or obtain a commercial license if Open Token exceeds US$20 million in monthly revenue.
DeepSeek
R1, V3.1, and V3.2 are offered under MIT terms. Copyright and license notices must be preserved.
Qwen 3
Apache 2.0: Qwen3 and Qwen3-VL
Qwen3, Qwen3 Coder, Qwen3 Next, and Qwen3 VL models require preservation of Apache 2.0 notices and any bundled attribution files.
MiniMax M2, M2.1, and M2.5
Modified MIT for M2 and M2.1; MiniMax Model License for M2.5
The product identifies MiniMax M2.1 and M2.5 by name. M2 attribution also becomes mandatory above 100 million monthly active users or US$30 million annual recurring revenue. For M2.5, Customer must preserve the license and attribution notice and follow its prohibited uses, including no military purpose and no fully automated adverse decision affecting legal rights or access to essential services.
Moonshot AI Kimi
Modified MIT for Kimi K2 Thinking and Kimi K2.5; Kimi K3 License
The catalog identifies Kimi models by name. Attribution becomes mandatory at 100 million monthly active users or US$20 million monthly revenue. The Kimi K3 Model-as-a-Service license requires a separate agreement once group revenue exceeds US$20 million over a rolling twelve-month period unless an express relay or certified-inference-partner exemption applies.
Z.AI GLM
Permissive MIT or Apache 2.0 terms, including the GLM-5 license
GLM 4.7, GLM 4.7 Flash, and GLM 5 require preservation of applicable copyright and license notices.
NVIDIA Nemotron
NVIDIA Nemotron Open Model License and NVIDIA Open Model License
Nemotron 3 Nano and Super require the license and notice to accompany permitted distribution. For Nemotron Nano 9B v2 and 12B v2 VL, rights can terminate if guardrails are bypassed. Customer must not disable safeguards and must follow NVIDIA's Trustworthy AI requirements.
4. BYOK-only families
The following catalog families are available only when Customer connects its own Amazon Bedrock account. Open Token does not grant additional rights to these models. Customer must ensure its AWS and provider agreements authorize the intended use.
Anthropic Claude
Anthropic on Bedrock commercial terms
Anthropic's Bedrock terms restrict resale without Anthropic approval.
OpenAI gpt-oss
Although the open-weight license is permissive, Open Token currently makes gpt-oss and safeguard variants BYOK-only as a platform policy. Customer must preserve Apache 2.0 notices and follow applicable use rules.
Writer Palmyra
Writer's terms restrict reselling or leasing direct access to its service.
xAI Grok
xAI's terms restrict resale, sublicensing, timesharing, and service-bureau use.
Mistral proprietary and research models
Mistral Large 24.02, Mistral Small 24.02, and Pixtral Large remain BYOK-only because public resale rights are absent or the upstream license is research-only.
5. Customer notices and downstream products
Customer is responsible for identifying the model used in its product when a license requires attribution, supplying applicable license and policy links to downstream users, and enforcing model use restrictions. Open Token's display of a notice does not replace Customer's own obligations when it redistributes Outputs, exposes model access, or incorporates a model into another service.
6. Corrections
Provider terms change and model cards can contain conflicting links. Report a missing or inaccurate license notice to support@getopentoken.com. We may pause a model while clarifying its terms.