Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- AWS announced on November 11, 2026 that Ornith-1.0-9B and Ornith-1.0-35B are available through SageMaker JumpStart.
- AWS positions 9B as a dense model for single-GPU coding workflows and lists a 256K-token context window. The announcement gives no minimum-memory requirement or device compatibility list.
- AWS says the 35B MoE model can run on a single 24GB consumer GPU. That local hardware statement is not a specification for an AWS hosted endpoint.
- The announcement gives no invocation IDs, Region eligibility, managed instance configuration or numeric price. Check the JumpStart model card and current billing details in the account you plan to use.
Confirm what JumpStart made available
AWS's release note is marked November 11, 2026 and announces that DeepReinforce AI's Ornith-1.0-9B and Ornith-1.0-35B are available through Amazon SageMaker JumpStart. It covers these two models; it does not establish that every account, Region or configuration can deploy them.
AWS describes Ornith 1.0 as an open-model family for agent workflows that use tools to complete coding tasks and says it is released under the MIT license. The model license and AWS hosting charges are separate matters.
Compare 9B and 35B against the coding task
Ornith-1.0-9B is a 9B dense model post-trained on Qwen 3.5 9B. AWS lists a 256K-token context window, positioning it for single-GPU coding workflows for individual developers and small teams. The announcement does not specify minimum memory, quantization or a device compatibility list.
Ornith-1.0-35B is a 35B mixture-of-experts (MoE) model post-trained on Qwen 3.5. AWS positions it as the stronger coding option and says it can run on one 24GB consumer GPU. This is AWS's model-deployment description, not a local compatibility test or a managed-instance specification.
Can single-GPU claims determine your device or AWS instance?
AWS's statements about 9B single-GPU deployment, including consumer laptops, and 35B deployment on one 24GB consumer GPU describe model deployment characteristics. They do not name a SageMaker JumpStart instance type, endpoint parameters, memory configuration or Region. Do not infer that a particular AWS instance can host either model, or extend the consumer-GPU claim to every local device.
For a local trial, first check whether your actual GPU and runtime can accommodate the model, then run a small evaluation on your own prompts, codebase and tool workflow. Do not infer compatibility from parameter count or the announcement's single-GPU statements alone.
What should you check before creating hosted resources?
For an AWS hosted deployment, open the JumpStart model catalog or model card in the account you intend to use. Check whether that account and Region can select the model, along with its exact invocation identifier, deployment parameters and resource requirements. SageMaker documentation describes JumpStart access through Studio, the console and the Python SDK, but the announcement does not provide a copy-ready model ID or invocation code.
Include account permissions, organization restrictions and resource quotas in the pre-creation checks; hold off on starting resources until these are clear.
How can you compare actual costs and coding results?
The SageMaker AI pricing page says real-time inference is charged for usage of the selected instance type. The Ornith announcement gives no hosting price for either model. Before deployment, check current pricing for the resources and Region selected, and establish how to stop or delete test resources and how that affects billing.
For a fair comparison, hold an approved set of representative coding tasks, prompts, tools and budget constant. Record task completion, whether outputs pass your acceptance checks, failure modes, elapsed time and actual charges. AWS's benchmark statement is vendor-reported and includes no scores or setup, so it cannot replace your evaluation. Set quality, latency and budget limits, plus conditions for stopping test resources.
Sources
Frequently Asked Questions
If Ornith is missing from my catalog, does that mean it has been withdrawn?
One account’s catalog result does not establish global availability. Check the target account and Region, current catalog and applicable permissions. The announcement provides no complete Region or account-eligibility list. Ask AWS about specific availability if it remains unclear; changing networks does not grant eligibility.
Do the MIT licenses make SageMaker hosted inference free?
No. The MIT license describes the model release. AWS hosting or compute resources may still incur charges. Check current SageMaker pricing for the resources and usage pattern you select.