← Back to feed

GPT Image 2 is now listed on OpenRouter at $8 per million input tokens and $8 per million output tokens, with a 400K context window and direct routing to OpenAI.

The pricing matters because it makes image generation legible to the same teams that already buy text, code, and multimodal inference by the token. Instead of a flat price per generated asset, OpenRouter’s listing frames GPT Image 2 as another meter in the API stack: context in, image work out, usage tracked alongside the rest of the model fleet.

What Is Listed

• Model: OpenAI GPT Image 2
• Route: one provider, forwarded directly to OpenAI
• Pricing: $8 per million input tokens, $8 per million output tokens
• Context: 400K
• Release date: June 24, 2026
• Use case: high-fidelity image generation and editing through the Images API

That is a different procurement story from the older image-generation market, where vendors pushed per-image pricing, quality tiers, and resolution surcharges. Token pricing makes usage easier to fold into agentic workflows, where the image step is only one tool call inside a larger chain.

Why Token Pricing Changes the Workflow

Design and marketing teams usually think in assets. Engineering teams think in budget controls, retries, context windows, latency, and observability. GPT Image 2 being sold through a model router pushes image generation closer to the engineering side of the house.

That matters for agents. A coding agent can generate UI mockups. A commerce agent can create product variants. A presentation agent can build diagrams. A support agent can edit screenshots. In all of those cases, image generation is not a separate creative suite. It is a callable capability inside a larger system.

The 400K context window is the tell. OpenAI is not positioning this only as prompt-to-picture. Long context makes room for brand systems, product specs, multi-page references, long edit instructions, and previous conversation state. The image model becomes a production component rather than a one-off creative endpoint.

The Competitive Pressure

OpenRouter already exposes multiple image models with different pricing structures, including flat per-image pricing and per-megapixel billing. GPT Image 2’s token-metered format gives buyers a cleaner comparison against the broader OpenAI stack, but it also makes cost estimation more sensitive to prompt size, reference context, and automated retries.

The practical question is not whether $8/$8 is cheap in isolation. It is whether a token-priced image model lets teams replace disconnected design tools with programmatic generation that can be routed, observed, budgeted, and embedded into agent workflows.

That is where image models are headed: less gallery, more infrastructure.