Vercel AI Updates: July 31, 2026

1. MiniMax H3 Video Model Arrives on AI Gateway

Vercel. Vercel added MiniMax H3, a video generation model, to its AI Gateway. The model produces 5- to 15-second mp4 clips at 2K resolution from a text prompt, a starting image, a pair of first and last keyframes, or multimodal reference material, and supports aspect ratios ranging from 21:9 to 9:16. Developers call it by setting the model identifier to minimax/minimax-h3 in the AI SDK. Source

2. Inkling Small From Thinking Machines Lands on AI Gateway

Vercel. Vercel made Inkling Small from Thinking Machines available on its AI Gateway. The model delivers performance comparable to the larger Inkling model at roughly a quarter of the size, with multimodal reasoning over audio and images, tool use, and adjustable thinking effort to trade output quality against speed and cost. It runs with Zero Data Retention and no markup on inference, and is accessed by setting the model to thinkingmachines/inkling-small. Source

3. AI Gateway Cuts GPT-5.6 Prices and Boosts Fast Mode

Vercel. Vercel updated GPT-5.6 pricing and speed on its AI Gateway. GPT-5.6 Luna dropped 80 percent to $0.2 per million input tokens on short context, GPT-5.6 Terra dropped 20 percent to $2 per million input tokens, and GPT-5.6 Sol’s fast mode increased from 1.5x to 2.5x faster with pricing unchanged. Vercel adds no markup on token pricing, so the new rates reach developers at the upstream price with no code changes required. Source