DeepSeek V4 Flash on Vercel's AI Gateway has been updated with new model weights, boosting its Terminal-Bench score from 56.9 to 82.7 — a 25.8-point improvement reflecting stronger agentic capabilities. Requests using the existing model ID `deepseek/deepseek-v4-flash` automatically pick up the new weights with no code changes required. Currently only DeepSeek serves the updated weights; other providers including Zero Data Retention options are coming next week. The model can be used via the AI SDK and is also configurable for coding agents through the `vercel ai-gateway coding-agents setup` command. AI Gateway passes through provider pricing with no markup or platform fee.
Questions this post answers
Does upgrading to the new DeepSeek V4 Flash weights on Vercel AI Gateway require changing my model ID or code?
No, requests to the deepseek/deepseek-v4-flash model ID automatically pick up the new weights with no change to the model ID or code required. The update improved the model's Terminal-Bench score from 56.9 to 82.7, reflecting notably stronger agentic capabilities, and DeepSeek is currently the only provider serving these updated weights. daily.dev keeps developers tracking model upgrades like this DeepSeek weight swap without breaking their pipelines.
How much did DeepSeek V4 Flash's agentic performance improve on Terminal-Bench with the new weights?
DeepSeek V4 Flash's Terminal-Bench score jumped 25.8 points, from 56.9 in the April preview to 82.7 with the updated weights served through Vercel AI Gateway. This reflects a substantial gain in agentic capability, relevant for tasks like fixing failing tests or opening pull requests autonomously. Developers comparing coding agent models can follow benchmark shifts like this one on daily.dev.
How do I connect DeepSeek V4 Flash to a coding agent through Vercel AI Gateway?
Run the vercel ai-gateway coding-agents setup command to connect agents to AI Gateway, then select deepseek/deepseek-v4-flash in the agent's model configuration. AI Gateway routes requests to providers serving the updated weights by default, with no markup on provider pricing and no platform fee on inference, including Bring Your Own Key requests. daily.dev helps developers wiring up coding agents stay current on gateway setup steps like this.
Share this post