DeepSeek V4.1 Flash lands on Vercel AI Gateway with vision and 1M-token context

What launched

Vercel has added DeepSeek V4.1 Flash to the models reachable through its AI Gateway, according to a changelog post on the Vercel blog dated September 10, 2026. The headline capability is vision: the model accepts text and images together in one request, so an application can hand it a screenshot and ask questions about it, interpret a chart, or extract information from other visual material.

A large window and even larger outputs

Two figures stand out in the specification. The context window sits at 1 million tokens, and the model can generate responses of up to 384,000 tokens. That pairing is uncommon — plenty of long-context models still cap output at a small fraction of what they can read — and it positions V4.1 Flash for work that involves both absorbing large inputs and producing extensive results, such as changes spanning a sizeable codebase or long document synthesis.

Reasoning, tool use, and prompt caching round out the feature set. Caching in particular can lower cost and latency for workloads that repeatedly send the same leading content, such as system prompts or reference documents.

An architecture split in two

Vercel's description also points to a structural change: the work of understanding input is separated from the work of producing output, using distinct components for each, which the company says cuts how much computation is active during either phase. The announcement ships without benchmark numbers, so developers comparing against their current models will need to run their own tests; Vercel's model playground is available for exactly that.

How to use it

On the Gateway, the model is addressed as deepseek/deepseek-v4.1-flash. Vercel notes that coding agents including Claude Code, Codex, and Cursor can be pointed at it: install the latest Vercel CLI, run its gateway setup command, and select the model inside the agent. The company documents the flow in its guide for coding agents.

What the Gateway layer adds

AI Gateway itself serves as a single entry point for calling many models, with usage and cost tracking, configurable retries and failover, and performance optimizations that Vercel claims push uptime above what providers offer directly. The platform also lists custom reporting, Zero Data Retention support, per-API-key budgets, and routing rules among its features. On cost, Vercel says the gateway passes provider rates through without a markup and takes no platform fee on inference, including on Bring Your Own Key requests.

Why it matters

A multimodal model with a 1 million token window and 384,000 token outputs is now one API call away for teams already on Vercel's infrastructure, at the same price as connecting to the provider directly. The practical draws are convenience — one interface, retries and failover handled centrally, spend controls per key — alongside a capability profile suited to tasks that mix images, long documents, and code. The split input and output processing may translate into real efficiency gains, though the announcement includes no measurements to confirm it. And as with any launch, the capability claims come from the platform announcing availability rather than independent evaluation, so how V4.1 Flash stacks up against peers will only become clear once outside benchmarks appear.