Your application. One endpoint.
We are preparing hosted language models with project-based token billing. Models and tariffs are not activated yet. You can review the planned lineup and integration interface in Console.
View Model API in Console →GLOBALATLAS AI · MODEL API
ATLAS Model API
ATLAS Model API is preparing for launch: hosted language models, separate input/output token pricing and project-based usage. Explore the planned integration.
Preparing for launchTransparent token accounting
Separate prices per million input and output tokens. Actual runtime usage, prepaid USD balance, per-project history and scoped API keys.
A model becomes callable after its endpoint, revision and tariff have been verified. Connection status and published prices are shown in Console.
Planned model lineup
Qwen2.5-7B-Instruct
Private text assistants and multilingual chat
Instruction-tuned text model. Document search requires a separate retrieval/indexing service; images and audio are not supported by this checkpoint.
Compare with a dedicated configuration →Qwen2.5-14B-Instruct
Business assistants and structured document processing
Instruction-tuned text model. Document search requires a separate retrieval/indexing service; images and audio are not supported by this checkpoint.
Compare with a dedicated configuration →Qwen2.5-32B-Instruct
Complex instructions and a larger self-hosted language model
Instruction-tuned text model. Document search requires a separate retrieval/indexing service; images and audio are not supported by this checkpoint.
Compare with a dedicated configuration →Mistral-7B-Instruct-v0.3
Text services and compact application backends
Instruction-tuned text model with the v3 tokenizer. Function calling needs a compatible chat template and application-side tool execution.
Compare with a dedicated configuration →DeepSeek-R1-Distill-Qwen-14B
Reasoning experiments and domain-specific assistants
A separate DeepSeek reasoning model distilled using DeepSeek-R1 outputs on a Qwen2.5-14B base. It is not the full 671B DeepSeek-R1 model or a bundle of two models.
Compare with a dedicated configuration →DeepSeek-R1-Distill-Qwen-32B
Larger reasoning workloads on dedicated infrastructure
A separate DeepSeek reasoning model distilled using DeepSeek-R1 outputs on a Qwen2.5-32B base. It is not the full 671B DeepSeek-R1 model or a bundle of two models.
Compare with a dedicated configuration →Dedicated server or token API?
Choose a dedicated configuration for runtime control and a fixed infrastructure budget. Model API suits variable request volumes; input and generated output are metered separately.