Claude Sonnet 5.5 arrives on Snowflake Cortex AI in public preview

What is available now and what is coming

At launch, Claude Sonnet 5.5 can be called through Cortex Inference. Developers reach it through their Snowflake account endpoint and can use the Anthropic SDK with Snowflake's Cortex endpoint. The example in Snowflake's post authenticates with a Bearer token and requests the model under the name claude-sonnet-5-5.

Support in other parts of the platform is not live yet. According to Snowflake, Snowflake CoWork, Snowflake CoCo, Cortex Agents and Cortex AI Functions are coming soon. The post also carries a forward-looking statement noting that such plans are not commitments to deliver any product offerings.

What Anthropic says about the model

Snowflake describes Claude Sonnet 5.5 as Anthropic's most capable Sonnet model yet and a clear upgrade over Sonnet 5 for well-scoped everyday tasks, bug fixes, and polished documents, slides and spreadsheets. The figures in the post are reported by Anthropic:

• A 70.6% score on Terminal-Bench 4.0, almost a 7x improvement over the 10.3% recorded for Sonnet 5.
• Performance nearly level with Opus 5.5 on GDPval-AA, a benchmark covering real-world work across 44 occupations.
• Substantial gains on chart recognition and computer use.
• Output generated more than 30% faster than Sonnet 5.

Anthropic positions Sonnet 5.5 as a faster complement to Opus 5.5. Opus 5.5 remains the better choice for complex, open-ended work that requires sustained judgment, while Sonnet 5.5 is described as optimal for well-scoped tasks where speed matters. Snowflake advises teams to evaluate Opus 5.5 for the more open-ended jobs.

Intended workloads and Snowflake products

Snowflake lists several workloads that fit the model: targeted coding tasks such as fixing bugs and understanding an unfamiliar codebase, producing documents, slide decks and spreadsheets (including following an existing template), everyday knowledge work, reading charts and other visual inputs, and latency-sensitive interactive workflows.

For Snowflake CoCo, the company's AI coding agent, Sonnet 5.5 is expected to bring improved reasoning, code generation and support for longer-running, agentic tasks across the CoCo desktop, the Snowsight interface, the CLI and local dev tools. Snowflake says the model is especially strong at turning a broad product idea into a working first version and refining it over several rounds of feedback.

For Snowflake CoWork, described as a personal work agent for knowledge workers, the post says the model is built for long-running agents that sustain multi-step tasks and check their own work. CoWork features named in the post include Deep Research, proactive automations, reusable User Skills, MCP connectors for Gmail, Jira, Slack and Salesforce, and a native mobile app.

For Cortex AI Functions, analysts will be able to use the AI_COMPLETE function to apply natural language instructions in SQL. Snowflake's example asks the model to compare a written claim, that revenue grew 25%, with underlying figures of $100 million and $120 million, and to flag which claims are supported, contradicted or unsupported.

What it means for Snowflake users

For now, the practical change is limited to applications built on Cortex Inference, where teams get a faster Claude option for everyday coding, analysis and document work without moving data outside Snowflake's governance boundary. Wider use inside CoWork, CoCo, Cortex Agents and SQL-based AI Functions depends on the follow-up releases Snowflake has flagged as coming soon, with no dates given in the post.