OpenRouter Alternative for Claude Code and GPT: Fees and Prices Compared (2026)
OpenRouter charges providers' list prices and adds a 5.5% fee when you buy credits ($0.80 minimum; 5% for crypto). OpenLux takes the opposite approach: it sells Claude, GPT and Gemini through discounted resource groups, where the cheapest Claude Code group is priced at about 18% of Anthropic's list price. This page, written by the OpenLux team, compares the two on fees, prices, compatibility and trade-offs, so you can decide which fits your workload. Prices were checked on November 20, 2026.
Quick answer
| OpenRouter | OpenLux | |
|---|---|---|
| Model price | Same as the provider's list price ("no markup on inference pricing") | List price × a group ratio. Claude Code groups: 0.18, 0.29 or 0.45 |
| Fee on top-ups | 5.5% on card purchases ($0.80 minimum), 5% on crypto; not refundable | None added |
| Model catalog | Very large, many providers | 300+ models: Claude, GPT, Gemini, DeepSeek, Qwen and more |
| API formats | OpenAI-compatible | OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, Gemini native |
| Works with Claude Code | Yes | Yes, through ANTHROPIC_BASE_URL |
| Bring your own key | Yes (free up to a monthly allowance, then 5%) | No |
| Minimum top-up | See OpenRouter | $1 |
If you need one bill across dozens of providers, BYOK, or provider routing controls, OpenRouter is built for that. If most of your spend is Claude Code, Codex or other Claude and GPT usage, and price is the main reason you are looking for an alternative, a discounted gateway like OpenLux costs a fraction of list price.
What you actually pay on OpenRouter
OpenRouter's FAQ states two things that decide the bill:
- Inference is passed through at the provider's price: you pay "the same rate as you would directly with the provider."
- Buying credits costs 5.5% by card, with a $0.80 minimum, or 5% with crypto. The fee is not refunded even if unused credits are.
So for Claude, OpenRouter is never cheaper than Anthropic's own API. It is list price plus about 5.5%. Its value is convenience, not discount.
What you pay on OpenLux
Every model on OpenLux has a base price equal to the provider's list price. Each resource group multiplies it by a ratio. These are the three Claude Code groups as listed on the pricing page on November 20, 2026:
| Group | Ratio | Group description (from the pricing page) |
|---|---|---|
| Claude-Code-1 | 0.1765 | Maximum discount, may queue at peak hours |
| Claude-Code-2 | 0.2941 | Moderate discount, moderate operational stability |
| Claude-Code-3 | 0.45 | Minor discount, excellent operational stability |
Per million tokens, that works out to:
| Model | Anthropic list (input / output) | Claude-Code-1 | Claude-Code-2 | Claude-Code-3 |
|---|---|---|---|---|
| Claude Opus 5.5 | $4 / $20 | $0.71 / $3.53 | $1.18 / $5.88 | $1.80 / $9.00 |
| Claude Sonnet 5.5 | $2 / $10 | $0.35 / $1.77 | $0.59 / $2.94 | $0.90 / $4.50 |
Cache reads follow the same rule. Opus 5.5 cache hits are $0.20 per million at list price, multiplied by the group ratio.
The trade-off is in the descriptions above. The cheapest group can queue at peak hours. If your work is latency-sensitive, such as a team waiting on Claude Code during the workday, the 0.45 group is the safer choice. It is still less than half of list price.
OpenLux also sells a direct Anthropic group at 1.1× list price. That is more expensive than OpenRouter, so it is left out of this comparison.
A worked example
Say a month of Claude Code work uses 10 million input tokens and 1 million output tokens on Claude Opus 5.5, ignoring caching to keep the arithmetic simple:
| Option | Cost |
|---|---|
| Anthropic API, list price | 10 × $4 + 1 × $20 = $60.00 |
| OpenRouter | $60 of credits + 5.5% fee = $63.30 |
| OpenLux Claude-Code-3 | $60 × 0.45 = $27.00 |
| OpenLux Claude-Code-2 | $60 × 0.2941 = $17.65 |
| OpenLux Claude-Code-1 | $60 × 0.1765 = $10.59 |
Real Claude Code sessions lean heavily on cache reads, which lowers every row. The ratios between rows stay the same.
Switching Claude Code to OpenLux
Claude Code reads two environment variables for an LLM gateway:
export ANTHROPIC_BASE_URL=https://api.openlux.ai
export ANTHROPIC_AUTH_TOKEN=<your OpenLux key>
- Do not add
/v1to the base URL. Claude Code appends/v1/messagesitself. If you add it, the request goes to/v1/v1/messagesand returns 404. - Create the key on the token page, and pick the group when you create it. The key is billed at that group's price.
- For OpenAI-compatible tools, such as SDKs, Cursor or Cline, use
https://api.openlux.ai/v1as the base URL. - To check that the key and URL are right, call
GET https://api.openlux.ai/v1/modelswith the key. A model list back means both are correct; a 401 means the key is wrong or missing.
Which one to choose
- Choose OpenRouter if you want BYOK, many niche providers in one account, or detailed provider routing, and you accept list price plus the credit fee.
- Choose OpenLux if your spend is mostly Claude Code, Codex, Claude or GPT, and you want to pay well under list price. Choose your group by how much peak-hour queueing you can tolerate.
- Stay on Anthropic directly if you need Anthropic's own contract, data terms or enterprise support.
Payment on OpenLux: bank cards, Apple Pay, Google Pay, Alipay, WeChat Pay and crypto.
See OpenLux plans and set up Claude Code →
FAQ
Does OpenRouter mark up model prices?
No. Its FAQ says inference is at the provider's rate. Its revenue is the 5.5% fee on card credit purchases ($0.80 minimum), or 5% on crypto.
Why is OpenLux cheaper than list price?
Each group is backed by a different upstream resource, and priced by its ratio. Cheaper groups come with weaker guarantees, which the pricing page states for each group, for example "may queue at peak hours." Prices change, so check the pricing page before you top up.
Can I use the same key for Claude and GPT?
A key is billed at the group you choose when you create it, and the pricing page lists which models each group carries. Claude Code groups carry Claude models and Codex groups carry GPT models, so most people create one key for each.