Snowflake Cortex AI review

The catalogue in claude opus 5, claude opus 4 8 and claude sonnet 5 mixes premium families from several vendors with open weights, and it includes embedding models alongside the text models. A single surface covers several providers, so a team can route jobs to different models without a client for each vendor.

The largest context window is 1,000,000 tokens, which puts long documents and wide result sets into a single request. Across the brands the site tracks, that figure ranks 21 of 72, a solid showing rather than the widest available.

A free tier is included. That lets a small team try queries against its own data before committing any budget, and it is the quickest way to see how the models in claude opus 5, claude opus 4 8 and claude sonnet 5 behave on your own workloads.

It is the AI layer of Snowflake Inc., founded in 2012 and headquartered in the United States. That matters for buyers who want model calls to sit next to the same warehouse and governance setup as the rest of their data.

Support runs through help centre and community, so answers come from the help centre documentation or the community rather than a direct call. For an outage on a paid workload that means more self serving, so teams with strict response expectations should weigh that before signing up.

Azure AI Foundry adds image, audio and realtime models alongside text in gpt 6 astra, gpt 6 luna and gpt 6 sol. Snowflake Cortex AI keeps the catalogue on text and embedding work, so teams needing media generation should look elsewhere.

Salad runs a short open model set, qwen3.6 35b a3b, qwen3.6 27b and qwen3.5 9b, which is easier to reason about than the wide mix Snowflake offers. Supascale AI takes the opposite approach, listing many provider families in DeepSeek V4 Pro, GPT 5.6 Sol and GPT 5.6 Terra for teams that want one gateway across vendors.