Models and regional availability for Cortex AI Functions¶

Snowflake Cortex AI Functions support a range of language and embedding models with varying capabilities and context-window sizes. Use this reference to choose a model for your workload, check its input and output limits, confirm which regions a function runs in, and understand legacy and end-of-life access.

Choosing a model¶

The Snowflake Cortex AI_COMPLETE function supports models from several providers, with varying capability, latency, and cost. To achieve the best performance per credit, choose a model that’s a good match for the content size and complexity of your task. The following sections group models by family. For each model’s input and output limits, see Model restrictions.

Large models¶

If you’re not sure where to start, try a large model first to establish a quality baseline, then evaluate smaller models against it.

• Anthropic Claude Opus (claude-opus-5-5, Public Preview; claude-opus-5): Anthropic’s most capable models, suited for complex reasoning, coding, long-horizon agentic work, and analysis of large document collections. Both models have a 1,000,000-token context window and up to 128,000 output tokens.
• Google Gemini (gemini-4-argon, Public Preview): Google’s most capable Gemini model in Snowflake Cortex, suited for advanced reasoning and multimodal tasks, with a 1,000,000-token context window.
• OpenAI GPT Astra (openai-gpt-6-astra, Public Preview): OpenAI’s most capable model, suited for complex, demanding reasoning work.
• OpenAI GPT Sol (openai-gpt-6-sol, openai-gpt-5.6-sol, Public Preview): OpenAI’s frontier models for agentic coding and long-horizon tasks.

Medium models¶

Medium models balance quality with lower latency and cost, and suit most production workloads.

• Anthropic Claude Sonnet (claude-sonnet-5-5, Public Preview; claude-sonnet-5): strong performance on coding, tool use, and agentic tasks at lower latency and cost than Claude Opus. claude-sonnet-5 has a 1,000,000-token context window and up to 64,000 output tokens.
• xAI Grok (grok-4.7, Public Preview): a general-purpose model for reasoning and agentic tasks, with a 500,000-token context window.
• OpenAI GPT Terra (openai-gpt-6.1-terra, openai-gpt-5.6-terra, Public Preview): balanced models for everyday production work, including code generation, content workflows, and structured data extraction. openai-gpt-5.6-terra has a 1,050,000-token context window and up to 128,000 output tokens.

Small models¶

Small models are fast and cost-efficient, and suit high-volume tasks like classification, extraction, and summarization.

• Anthropic Claude Haiku (claude-haiku-4-5): optimized for low-latency, high-throughput workloads, with a 200,000-token context window and up to 64,000 output tokens.
• Google Gemini Flash (gemini-3.8-flash): a fast, low-cost Gemini model for high-volume tasks, with a 1,000,000-token context window.
• OpenAI GPT Luna (openai-gpt-6-luna, openai-gpt-5.6-luna, Public Preview): fast, low-cost models for repeatable work at scale, such as summarizing documents, extracting information, and answering focused questions. Both have a 1,050,000-token context window and up to 128,000 output tokens.
• OpenAI GPT mini and nano (openai-gpt-5.4-mini, openai-gpt-5.4-nano): small, fast GPT models, each with a 400,000-token context window and up to 128,000 output tokens.

Many of these models are available only through cross-region inference. Check Model availability for your region before you choose.

Function availability by region¶

Snowflake Cortex AI functions are available in the following regions. If your region is not listed for a particular function, use cross-region inference.

The tables in this section cover the functions themselves. The models that AI_COMPLETE and AI_EMBED accept have their own regional availability, documented in Model availability. A function being available in your region does not mean every model it accepts is, so check both.

† Functions are available to Azure US FedRAMP High Plus accounts through cross-region inference with processing on AWS in the United States. See US FedRAMP High and DoD gov regions.

The following Snowflake Cortex AI functions are available in the following extended regions.

Model restrictions¶

Models used by Snowflake Cortex have limitations on size as described in the table below. Sizes are given in tokens. According to industry estimates, tokens generally represent about four characters of text, so the number of words corresponding to a token limit is less than the number of tokens. Inputs exceeding the context window limit result in an error. Output that exceeds the context window limit is truncated.

The maximum size of the output that a model can produce is limited by the following:

• The model’s output token limit.
• The space available in the context window after the model consumes the input tokens.

For example, claude-sonnet-4-6 has a context window of 1,000,000 tokens. If 100,000 tokens are used for the input, the model can generate up to 8,192 tokens. However, if 195,000 tokens are used as input, then the model can only generate up to 5,000 tokens for a total of 200,000 tokens.

Model lifecycle: legacy and end-of-life¶

Snowflake retires Cortex models in two stages: legacy, then end-of-life. A model marked legacy is not yet unavailable to every account. Access depends on whether your account already used that model.

Legacy¶

After a model’s legacy date:

• If your account used the model anywhere before the legacy date, the account can continue to use it in every supported application until the model’s end-of-life date. Supported applications include AI_COMPLETE, CORTEX.COMPLETE, the Agents API, Cortex Inference, and Snowflake CoWork.
• If your account had not used the model anywhere before the legacy date, the account can’t start using it. Queries and API calls that name the model fail.

Any usage of the model in the account before the legacy date means that the account can continue to use the model until end-of-life, including in applications that had not called that model before.

Snowflake recommends that you migrate to a supported model before the end-of-life date, even if your account can still call the legacy model.

End-of-life¶

After a model’s end-of-life date, the model is no longer available to any account, regardless of prior usage. Queries and API calls that name the model fail.

Use SHOW CORTEX BASE MODELS to see each model’s lifecycle_status, legacy_date, and eol_date. Deprecation dates for specific models are announced in unbundled behavior change releases, such as Cortex model deprecations for August 2026.