Are resold API keys cheaper than going direct?

Often yes on a per-token basis, because gateways buy volume and resell at a markup or discount. Passthrough gateways match the lab price and add a small credit fee; discount gateways claim 20 to 60 percent below list on a smaller catalogue. Verify the live rate and the provider's terms before sending production traffic.

AI model price comparison (2026)

Q: Which gateway is cheapest for Claude Opus 4.8?

Against a $5 / $25 per 1M official list price, Novarelay and CometAPI both charge about $4 / $20 (20% below list), OpenRouter passes the list rate through (0% model markup, plus a 5.5% credit-purchase fee), and AiHubMix varies by model with a free tier. Compare the absolute price per 1M tokens after any fees before you build.

Q: Why do the discount percentages differ between gateways?

Each gateway sets its own baseline and its own model catalogue, so a headline discount only means something against the official list price. Always compare the absolute price per 1M tokens, not just the percentage.

Official list price vs. each gateway’s price, per 1M tokens · Last updated July 2026. Figures are taken from each provider’s own pricing page or its stated discount — always verify on the provider’s site before you build.

The tables below compare four AI API gateways — Novarelay, OpenRouter, CometAPI and AiHubMix — against the model vendor’s official list price. For every model you get the exact input and output price per 1M tokens, plus the discount each gateway applies. Prices are in US dollars, taxes excluded.

How to read this: a discount only means something against a baseline, and gateways quote different baselines and different catalogues. Compare the absolute price per 1M tokens (and any credit-purchase fee), not just the percentage.

Get a Novarelay API key →

Headline pricing model by gateway

Gateway	Pricing model	Headline discount	Free tier	Top-up
Novarelay	Discounted resale	Up to 60% off list (varies by model)	No — pay-as-you-go	Credits from $1
OpenRouter	Passthrough	0% on models · +5.5% on credit purchases	Yes — free models	Credits (never expire)
CometAPI	Discounted resale	20% below list on all models	Trial credits	Pay-as-you-go, no minimum
AiHubMix	Mixed	Varies by model · 27+ models free	Yes — 27+ free, no card	Pay-as-you-go

Price per 1M tokens, model by model

Claude Opus 4.8

Official list price: $5.00 input / $25.00 output per 1M tokens (Anthropic).

Gateway	Input / 1M	Output / 1M	Discount vs list
Novarelay	$4.00	$20.00	−20%
CometAPI	$4.00	$20.00 †	−20%
OpenRouter	$5.00	$25.00	0% · +5.5% credits
AiHubMix	Varies · live rate in dashboard		Varies

Claude Sonnet 5

Official list price: $3.00 input / $15.00 output per 1M tokens (Anthropic).

Gateway	Input / 1M	Output / 1M	Discount vs list
Novarelay	$1.80	$9.00	−40%
CometAPI	$1.60 §	$12.00 †	−20%
OpenRouter	$3.00	$15.00	0% · +5.5% credits
AiHubMix	Varies · live rate in dashboard		Varies

Claude Fable 5

Official list price: $10.00 input / $50.00 output per 1M tokens (Anthropic).

Gateway	Input / 1M	Output / 1M	Discount vs list
Novarelay	$6.00	$30.00	−40%
CometAPI	$8.00	$40.00 †	−20%
OpenRouter	$10.00	$50.00	0% · +5.5% credits
AiHubMix	Varies · live rate in dashboard		Varies

GPT-5.5

Official list price: $5.00 input / $30.00 output per 1M tokens (OpenAI).

Gateway	Input / 1M	Output / 1M	Discount vs list
Novarelay	$2.00	$12.00	−60%
OpenRouter	$5.00	$30.00	0% · +5.5% credits
CometAPI	Lists GPT-5.6 at $60.00 / 1M ¶		−20%
AiHubMix	Varies · live rate in dashboard		Varies

Gemini 3.5 Flash

Official list price: $1.50 input / $9.00 output per 1M tokens (Google).

Gateway	Input / 1M	Output / 1M	Discount vs list
Novarelay	$1.20	$7.20	−20%
CometAPI	$1.20	$7.20 †	−20%
OpenRouter	$1.50	$9.00	0% · +5.5% credits
AiHubMix	Varies · live rate in dashboard		Varies

Gemini 3.1 Pro

Official list price: $2.00 input / $12.00 output per 1M tokens (Google).

Gateway	Input / 1M	Output / 1M	Discount vs list
Novarelay	$1.60	$9.60	−20%
CometAPI	$1.60	$9.60 †	−20%
OpenRouter	$2.00	$12.00	0% · +5.5% credits
AiHubMix	Varies · live rate in dashboard		Varies

GPT-Image-2

Official list price: $8.00 input / $30.00 output per 1M tokens (OpenAI). Some gateways bill this model per image (~$5 / image) instead of per token.

Gateway	Input / 1M	Output / 1M	Discount vs list
Novarelay	$6.40	$24.00	−20%
CometAPI	~$4.00 / image (vs $5 list)		−20%
OpenRouter	$8.00	$30.00	0% · +5.5% credits
AiHubMix	Varies · live rate in dashboard		Varies

† CometAPI publishes a per-model input rate and states every model is priced at least 20% below the provider’s official rate with no platform surcharge; the output figures shown apply that 20% to the vendor list price.
§ CometAPI quotes a $2.00 / 1M list for Claude Sonnet 5 and prices it at $1.60 (20% below its own baseline); we show it against the $3.00 vendor list for a like-for-like view — compare the absolute price.
¶ CometAPI’s current OpenAI flagship is GPT-5.6 at $60.00 / 1M (20% below its $75 list); it does not list GPT-5.5.
OpenRouter is passthrough — the model price matches the vendor list; the platform adds about 5.5% on credit top-ups instead of a per-token markup.
AiHubMix does not publish a single blanket discount; it prices per model (many near list) and offers 27+ models free. Check its dashboard for the live rate.

How we compare gateways

We look at five things that actually affect a build: model coverage (which models, how current), real price per 1M tokens (token rates vs list, plus any credit-purchase or platform fee), reliability (uptime, failover, a public status page), OpenAI compatibility (point an existing SDK at a new base URL), and developer experience (docs, dashboard, logs, support). We are an independent, affiliate-funded comparison — see our disclaimer.

Frequently asked questions

Which gateway is cheapest for Claude Opus 4.8?

Against a $5.00 / $25.00 per 1M list price, Novarelay and CometAPI both land around $4.00 / $20.00 (20% below list), OpenRouter passes the list rate through (0% model markup plus a ~5.5% credit-purchase fee), and AiHubMix varies by model with a free tier. For steady, high-volume traffic, compare the effective per-1M rate after fees and caching against buying an API key directly.

Why do the discount percentages differ between gateways?

Each gateway sets its own baseline and its own catalogue, so a headline percentage only means something against the official list price. For example, a gateway showing “20% off” on its own lower baseline can still be a higher absolute price than a competitor showing “40% off” the vendor list. Always compare the absolute price per 1M tokens.

Can I use one API key for GPT, Claude and Gemini?

Yes — that is the point of a gateway. A single OpenAI-compatible key reaches the models above from one endpoint and one balance, and you switch models by changing the model name.

Last updated: July 2026 · Independent comparison, funded by referral commissions. Model names and trademarks (OpenAI, GPT, Anthropic, Claude, Google, Gemini) belong to their respective owners and are used for identification only. Prices are set by each provider and can change — always verify before you build.