OpenRouter Drops Image API, Analytics, and Search Benchmarks

OpenRouter, an AI gateway startup acquired by Stripe for $7 billion on August 16, released three new tools within 72 hours: an Image API supporting 30+ models from eight providers, an analytics dashboard and API for tracking AI spend, and web search leaderboards benchmarking eight providers. The Image API normalizes parameters across providers and includes Zero Completion Insurance, while the analytics tools use a read-only management key. These are the first product moves under Stripe ownership, though no pricing changes have been announced.

OpenRouter shipped three tools across 72 hours this week. Each fixes a real integration problem. And because Stripe closed a $7 billion acquisition of the company on August 16, these are also the first product moves under new ownership — which makes the timing worth noting. The Acquisition Context OpenRouter’s pitch to developers has always been vendor neutrality: one API key, no preference for which model wins, routing based on cost and capability. That story was credible when the company was independent. Under Stripe, the dynamics are less clear — even if nothing has changed yet. No pricing changes have been announced. But the neutrality narrative now comes with an asterisk. The TechCrunch report on the acquisition https://techcrunch.com/2026/08/16/stripe-will-reportedly-acquire-ai-gateway-startup-openrouter-for-7b/ noted the $7B price tag is a 5.4x markup over OpenRouter’s Series B valuation from just three months prior. Keep that in mind as you evaluate how deeply to embed these new APIs into your stack. OpenRouter Image API: One Endpoint, 30+ Models Before this week, using multiple image generation providers meant integrating separate endpoints, auth patterns, and billing units. FLUX charged per megapixel. GPT Image billed per token. Seedream returned base64. You wrote the plumbing for each. That’s over now — at least if you’re already on OpenRouter. The new Image API lives at a single endpoint. Thirty-plus models, eight providers: FLUX, GPT Image, Gemini image family, Grok Imagine, Seedream, MAI-Image, Recraft, and Riverflow. The API normalizes parameters across all of them — resolution, aspect ratio, quality, output format, background transparency, and input references. POST https://openrouter.ai/api/v1/images Authorization: Bearer $OPENROUTER API KEY { "model": "black-forest-labs/flux.2-pro", "prompt": "your prompt here" } Providers with supports streaming: true can stream partial images over SSE — set "stream": true in your request. Currently, that means OpenAI-backed providers. Failed generations aren’t billed; OpenRouter calls this Zero Completion Insurance. The capability discovery endpoint is the part most will overlook: GET https://openrouter.ai/api/v1/images/models It returns every image model with its supported parameters per endpoint. You don’t have to read eight different provider docs to know what each model accepts. Full documentation is at the OpenRouter Image API announcement https://openrouter.ai/blog/announcements/image-api/ . Analytics Dashboard and API: See Where Your AI Budget Goes OpenRouter now shows what each model is costing your team, broken down by model, API key, and user or app. The dashboard includes a spend graph day, week, or month , a cost-by-model breakdown, and the ability to click any bar to drill into the underlying request logs. The terminal-accessible version is more useful for engineering teams. The Analytics API uses a management key — separate from your inference key, created at Settings → Management Keys. Management keys cannot make model calls, so they’re read-only and free to run. What they return: spend per model, OpenRouter credits vs. BYOK spend, request volume, token breakdown prompt vs. completion , and prompt cache utilization. Full setup is in the Analytics API cookbook https://openrouter.ai/docs/cookbook/administration/analytics-cost-control . Pull that data into your LLMOps dashboard or run it against your weekly model spend from a script. Web Search Leaderboards: Stop Guessing Which Engine to Use If you’re using web search in your agents, you’ve been guessing. OpenRouter benchmarked eight providers — Exa, Parallel, Perplexity, Brave, Tavily, Linkup, Firecrawl, You.com — across four task suites. The live results are at openrouter.ai/benchmarks/widesearch https://openrouter.ai/benchmarks/widesearch . A few findings worth surfacing: Exa Instant hit 306ms p50 latency on SealQA, running 3x faster than Parallel basic with 2.5x higher accuracy. Perplexity returned errors on 115 of 200 SimpleQA queries. The leaderboard grades four variables simultaneously — engine, model, pre-search vs. tool-call method, and search budget per request. OpenRouter’s recommendation: pick the task suite closest to your use case, find the cheapest config within a few points of the top score, then run your own evaluation set against configs that cost a bit more to see if the quality gap appears in your results. What the Pattern Signals Three releases in three days — Image API, Analytics, Search Leaderboards — each adds a Stripe-like primitive: unified billing visibility, capability discovery, and a quality comparison layer. OpenRouter is building infrastructure, not a convenience wrapper. The company’s “every modality, one API” direction https://openrouter.ai/blog/insights/every-modality-one-api/ suggests video and audio unified endpoints are coming next. Whether this is good for developers depends on whether the vendor-neutral promise survives the acquisition intact. For now, the tools are useful. Use them. But watch the pricing page.