Vercel adds free Ling 3.0 Flash Fin model and Cursor agent support to AI developer stack

Vercel shipped two developer-facing updates on August 27: a finance-tuned language model from Inclusion AI is now available at no cost through its AI Gateway, and the AI SDK harness layer gained an official adapter for the Cursor coding agent.

A finance-focused model on AI Gateway

According to the Vercel blog, Ling 3.0 Flash Fin is a finance-specialized variant of Inclusion AI's Ling 3.0 Flash. It offers a 256,000-token context window, can generate up to 32,000 output tokens, and supports both reasoning and function calling. Vercel positions it for financial research and analysis, particularly multi-step jobs that chain a series of tool calls together before returning a final answer.

Access is free through September 25 under the model name inclusionai/ling-3.0-flash-fin. There is a billing catch worth noting: requests to that standard name will begin accruing charges automatically once the promotional period ends. To guarantee no surprise bills, Vercel says developers can append a -free suffix to the model name (inclusionai/ling-3.0-flash-fin-free), which makes the model simply stop serving after the deadline instead of billing.

The model can also be wired into coding agents. Vercel's changelog points to its coding agents guide and a setup command, vercel ai-gateway coding-agents setup, which connects agents such as Claude Code, Codex, OpenCode, Cursor and Pi so the model can be selected from inside the agent.

For context, AI Gateway is Vercel's routing layer for model calls: a single API that handles usage and cost tracking, retries, failover and performance optimizations aimed at higher uptime than individual providers. Vercel says the gateway passes provider pricing through with no markup and charges no platform fee on inference, including on bring-your-own-key requests. Additional capabilities include Zero Data Retention support, budgets for API keys, routing rules and custom reporting.

Cursor joins the AI SDK harness layer

In a second announcement the same day, Vercel added Cursor to the AI SDK's harness layer through an official adapter, @ai-sdk/harness-cursor. The harness layer lets an application drive different coding agents behind one HarnessAgent interface, so teams can swap agents without modifying application code. Instantiating the agent with cursor as the harness is all that is required, per the Vercel blog.

Under the hood, the adapter builds on @ai-sdk/harness-acp and talks to Cursor via the Agent Client Protocol (ACP). Cursor joins a supported roster that already included Claude Code, Cline, Codex, Deep Agents, Grok Build, OpenCode and Pi, and Vercel says additional harnesses are coming.

Why it matters

Both changes reduce friction for teams building AI applications on Vercel's stack. A free, long-context model aimed at financial analysis gives developers a low-risk way to prototype domain-specific agents, especially ones that orchestrate many tool calls in sequence. The -free suffix detail is the practical takeaway for anyone who wants to experiment without committing to paid usage after September 25.

The Cursor adapter, meanwhile, moves the AI SDK toward being an agent-neutral abstraction. If a single interface can run Cursor, Claude Code, Codex, OpenCode or others, then choosing or changing an agent stops being an architectural decision and becomes a configuration option. Combined with AI Gateway's routing, cost controls and BYOK support, Vercel is steadily consolidating model access and agent execution into one layer of its platform.