Arctic Models by Snowflake
Details
Capabilities
About
Snowflake's Arctic family of LLMs is a powerful suite of large language models engineered for enterprise applications, emphasizing both intelligence and resource efficiency. Leading the collection is the Arctic model, which features a unique Mixture-of-Experts (MoE) hybrid transformer architecture. This combines a dense 10B parameter transformer with a 128x3.66B MoE MLP, resulting in an impressive total of 480B parameters, though only 17B are active at any given time. This innovative design allows Arctic to deliver high-quality outcomes while optimizing resource use, outperforming many other prominent open models in various benchmarks. The Arctic lineup also includes models like Arctic-Instruct, tailored for following instructions and generating high-quality responses from natural language queries, and Arctic Embed, a set of text embedding models engineered for retrieval tasks. All models are available under the Apache 2.0 license, fostering open usage and collaboration 12.
Current Variants
Use-when guidance is derived from seed capabilities, context, release, and replacement fields.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Arctic-TILT | Use when the workload needs 800M parameters. | 2024-05 | 800M parameters | Current |
Release Timeline
2 release groupsSpecifications(2 models)
| Model | Released | Context | Parameters | Structured Outputs |
|---|---|---|---|---|
| Arctic-TILT | 2024-05 | — | 800M | No |
Available From(4 providers)
Pricing
Frequently Asked Questions
- What is Arctic used for?
- Arctic is used for structured outputs and coding. The family description and listed model capabilities point to those workloads as the best fit.
- How does Arctic compare to Claude 3?
- Arctic by Snowflake is strongest where you need structured outputs, while Claude 3 by Anthropic is the closest related family to check for vision and multimodal work. Arctic has 2 listed variants and reaches up to 4k context, while Claude 3 reaches up to 200k context, so compare the specs and pricing tables before choosing a production model.
- Which Arctic model should I use?
- For the lowest listed input price, start with Arctic through Replicate API at $0.65/1M input tokens. For the most capable/latest local choice, evaluate Arctic-TILT.