Vercel Opens a Medical Model Free Until 4 October: Two Model IDs, One Starts Billing and One Just Stops

Vercel announced on 2026-09-04 that inclusionAI’s Ling 3.0 Flash Sante is on AI Gateway and free to use through 4 October.

The part worth remembering is not the free window. It is that there are two model IDs, and they behave in opposite ways when the offer ends. Verified 2026-09-08.

1. The two model IDs (official wording)

In practice:

• If you are evaluating and do not want to roll into paid usage, use the -free ID. When it ends you get an error, not an invoice.
• If you already know you will keep using it, the standard ID transitions seamlessly into billing.

Vercel also notes that free requests still appear in your spend dashboard and carry a trace — they just cost nothing. A dashboard entry is not a charge.

This design deserves copying. On most platforms a free period simply becomes a charge. Making “keep me on free” a distinct model ID is the friendlier option.

2. What the model is

Per the announcement:

• A health and medicine focused version of Ling 3.0 Flash
• Mixture-of-Experts, 124B total parameters, about 5.1B active per token
• 256K token context window
• Function calling supported
• Built for medical reasoning, professional healthcare tasks, deep research, evidence-based retrieval and multi-step medical workflows
• Retains the base model’s general reasoning, coding and agentic capabilities

3. Three caveats

1. “Medical” describes training and evaluation focus, not regulatory clearance. Any health-related use needs a qualified human in the loop. The announcement does not say this; we are saying it explicitly.
2. The window is one month — 4 September to 4 October. If your evaluation needs a quarter to show results, this is not long enough.
3. Free is not costless. Your request volume, data handling and integration hours remain. What this month is genuinely for is answering “is this model right for my use case”, not wiring it into production.

4. How to run a useful evaluation inside the window

Our suggested order:

1. Start with the -free ID to remove the risk of accidental billing.
2. Prepare 20–30 questions you actually face, not generic benchmarks. A specialised model only shows its value on specialised work.
3. Run a control group — your current model, same questions, same prompts. Evaluations without a control always look good.
4. Log latency and failure rates, not just answer quality. Whether 256K of context is usable under your real load is an empirical question.

Free-tier rules on other platforms: OpenRouter free model daily limits.

Source read directly on 2026-09-08: Vercel’s changelog entry Ling 3.0 Flash Sante is now available on AI Gateway for free (2026-09-04). Model ID behaviour, parameter counts and context length are official wording. Sections 3 and 4 are our own commentary, not vendor guidance; medical use is governed by your jurisdiction and professional judgement.