Architecture cost review
OpenRouter Pricing in October 2026: What You Pay on Top of Tokens
September changed OpenRouter's cost picture more than any month this year: a self-serve Business plan, in-region routing, a Batch API, a hosted shell and a new cache-aware router. Here is what each line costs.
Direct answer
OpenRouter passes through provider token prices and earns a fee on the credits you buy: 5.5% on Standard and 8% on the Business plan launched in September 2026. Business adds US and EU in-region routing and 1,000 workspaces, with no monthly minimum. The September Batch API cuts most per-token prices by 50% for work that can wait 24 hours, which outweighs the credit fee for batchable jobs.
Decision summary
| Decision area | What matters |
|---|---|
| Standard | Provider token rates + 5.5% fee on credit purchases |
| Business (Sept 2026) | Provider rates + 8% credit fee · US/EU in-region routing · 1,000 workspaces · no minimum |
| BYOK | First 1M requests a month free, then 5% of what the model would normally cost |
| Batch API | Typically 50% of per-token price, 24-hour window; xAI models 20% off |
| Hosted shell | $0.0001 per active sandbox second, billed in the request (beta) |
| Routers | openrouter/auto and typesafe/jev-router add no fee — you pay the routed model |
How OpenRouter makes money
OpenRouter charges the provider's token price for each model. Claude Opus 5.5 is $4 per million input and $20 per million output, GPT-6.1 Sol is $2 and $10, and GPT-6 Luna is $0.10 and $0.50. The margin is a fee on the credits you buy. On the Standard pay-as-you-go plan that fee is 5.5%, so $1,000 of model usage needs about $1,055 of credits.
Bring-your-own-key works differently. If you attach a provider key, the provider bills you for inference. OpenRouter charges nothing for the first million BYOK requests a month and 5% of the model's normal cost after that. BYOK suits teams with committed-spend discounts or existing provider contracts.
The new Business plan
In September OpenRouter launched a self-serve Business plan that any account or organisation can switch on in settings. It charges 8% on credit purchases instead of 5.5%. It still bills inference at provider rates, with no monthly minimum or contract. It raises the workspace cap from 5 to 1,000 and includes US and EU in-region routing.
On $10,000 of monthly tokens, the extra 2.5 points cost about $250. That is worth it if you need data residency, separate workspaces per client or team, or the governance that comes with them. If you do not, Standard is cheaper.
In-region routing and what it disables
Point a request at us.openrouter.ai or eu.openrouter.ai and OpenRouter decrypts, routes and serves it inside that region, using the same key and model IDs. If no in-region provider can serve the model, the request returns a 404 rather than silently leaving the region. Guardrails can restrict a key, member or workspace to allowed regions.
The cost catch is routing. openrouter/auto, Jev Router and other router models are not available on the regional domains. Regional traffic must name a fixed model, so the savings a router would find are not available there.
Batch API: the biggest September saving
The Batch API accepts an inline array of requests for one model and one endpoint shape. It supports chat completions, Responses, Anthropic Messages or embeddings, and returns results within 24 hours. Batch requests are typically billed at 50% of the model's per-token price. xAI models are 20% off in batch. Web search and prompt caching bill at their usual rates, and batches are text-only.
Each model with batch support lists a :batch variant. GPT-6.1 Sol drops from $2 and $10 to $1 and $5, Claude Opus 5.5 from $4 and $20 to $2 and $10, and GPT-6 Astra and Claude Fable 5.1 from $10 and $50 to $5 and $25. For overnight classification, evaluation runs and back-office summarisation, batch saves far more than the credit fee costs.
Server tools, routers and other line items
The hosted shell lets any tool-calling model run commands in a sandboxed Linux container. It uses openrouter:shell, or openrouter:bash for the Anthropic tool shape. Sandbox time costs $0.0001 per active second, billed inside the request, so ten minutes of active execution is $0.06. The Files API stores up to 10 GiB per workspace, and both tools are in beta.
Routers add no fee of their own. The Auto Router now takes a cost_tier band from low to max, and the new Jev Router chooses model and reasoning effort per turn while protecting the prompt cache. In both cases you pay the rates of whichever model serves the request.
September model prices worth knowing
Claude Opus 5.5, Claude Fable 5.1 and GPT-6 Astra took the top three places on OpenRouter's Intelligence Index in September. Opus 5.5 is the cheapest of the three at $4 and $20, against $10 and $50 for the other two. Below them, GPT-6.1 Sol is the current Sol release at $2 and $10, with higher rates above 272,000 prompt tokens. Grok 4.7 lists at $2 and $6.
At the cheap end, GPT-6 Luna costs $0.10 and $0.50. Gemini 3.8 Flash is listed at $0.75 and $3.75 after a 50% promotion, and Inception's Mercury 2.5 diffusion model at $0.04 and $0.15 after an 80% promotion. DeepSeek V4.1 Flash launched at $0.30 and $1.20, with off-peak hours at half price. Promotions end, so price production plans at the undiscounted rate.
- Claude Opus 5.5: $4 / $20 · batch $2 / $10
- GPT-6.1 Sol: $2 / $10 · batch $1 / $5
- GPT-6 Astra and Claude Fable 5.1: $10 / $50 · batch $5 / $25
- GPT-6 Luna: $0.10 / $0.50 · batch $0.05 / $0.25
- Grok 4.7: $2 / $6
Controls that stop surprise bills
Workspace budgets are now available on every plan. Give each product, team or client its own workspace, set a budget, and spending stops or alerts at the limit you choose. The new Analytics API exposes the Activity dashboard's data to scripts through a management key, so finance can reconcile spend by workspace and model without logging in.
Guardrails restrict which models, providers and regions a key, member or workspace may use. Combine them with provider.max_price on routed requests, and with explicit pools on Jev Router, so a router never resolves to a model you have not priced.
Key takeaways
- →OpenRouter passes through provider token prices and earns a fee on the credits you buy: 5.5% on Standard and 8% on the Business plan launched in September 2026. Business adds US and EU in-region routing and 1,000 workspaces, with no monthly minimum. The September Batch API cuts most per-token prices by 50% for work that can wait 24 hours, which outweighs the credit fee for batchable jobs.
- →Stay on Standard unless you need in-region routing or more than five workspaces, push every delay-tolerant job to the Batch API, and set workspace budgets before you scale.
- →Fees, discounts and model prices change; promotional discounts such as Gemini 3.8 Flash at 50% off and Mercury 2.5 at 80% off can end without much notice.
How this page was prepared
This September 2026 cluster uses OpenRouter's live models API, OpenRouter and TypeSafe documentation, and the Laya model cards, all checked on 1 October 2026. Vendor benchmark claims are attributed, third-party benchmarks are labelled as such, and every cost example states its token assumptions. We did not run a private benchmark for these pages.
Frequently asked questions
How much does OpenRouter charge?
OpenRouter passes through model token prices and charges 5.5% on credit purchases on the Standard plan, or 8% on the Business plan. BYOK is free for the first million requests a month, then 5% of the model's normal cost.
What is the OpenRouter Business plan?
A self-serve plan launched in September 2026 with an 8% credit fee, US and EU in-region routing, up to 1,000 workspaces and no monthly minimum or contract.
Does OpenRouter have a batch discount?
Yes. The Batch API, launched in September 2026, typically bills 50% of the per-token price for requests completed within 24 hours; xAI models are 20% off.
Can I use the Auto Router with EU data residency?
No. Router models, including openrouter/auto and Jev Router, are not available on the US or EU regional domains.