API aggregators: CometAPI, NanoGPT, OpenRouter and what else, and when one is worth it
Every price, term and referral rule was read on the vendor's own page: the pay-as-you-go aggregators and OpenCode Go on 19 November 2026, OpenInference on 20 November 2026. Three links are referral links - a qualifying CometAPI top-up may earn us a commission, eligible NanoGPT usage and a first Z.ai subscription may earn us credits - at no extra cost to you. Every other link goes to the vendor and earns us nothing.

An aggregator sells two things: one key instead of ten, and one invoice instead of ten. Only one service here publishes a rate below the vendor's list price.
CometAPI prices the models it lists at 0.8 of official, NanoGPT publishes list prices with no markup, and OpenRouter passes the provider rate through and charges on the purchase. The flat-rate gateways sell a monthly budget of usage instead, and two of them restrict it to coding agents.
What each aggregator actually sells
| Service | The offer | What it is really worth | Published margin or commission | Window and who qualifies |
|---|---|---|---|---|
| CometAPI: 500+ models at 0.8 of official | One key, one invoice; per token, per image, per second | The only below-list rate here: customer price = official x 0.8. Credits roll over | 20 % below official on the models it prices; 20 % of every referred top-up | Black Friday 24 Nov to 3 Dec 2026, 14 models, no code |
| NanoGPT: 1,263 models at vendor list price | OpenAI-compatible API at list price; $0 a month | Vendor price for text models - GPT-5.5 at $5/$30 per million tokens, OpenAI's own rate. Pro at $12 adds 60M input tokens a week and 5 % off eligible text models | 0 % markup; +5 % to pin a provider; bring-your-own-key at 5 % of the model cost. Referral: 10 % of eligible usage and 5 % of subscriptions to the inviter | Weekly reset Monday 00:00 UTC, no rollover; closed models and media bill from balance |
| OpenRouter: one key across many providers | One key and one balance; provider rates passed through | Provider list price plus a charge on buying credits | 5.5 % on credit purchases, $0.80 minimum. No publisher affiliate programme; its startup scheme needs a partner key | Purchases between $5 and $25,000; promo credits have no cash value |
| Openference: flat gateway over 56 models | One OpenAI-compatible endpoint with format conversion | Minion at $20 a month buys $60 of usage (3x), Boss at $80 buys $160 (2x). Until 5 December 2026 it rises to $100 and $300 | No markup on catalog rates; 5 % processing fee per on-demand top-up; no affiliate programme | 5-hour window 20 % of the budget, weekly 50 %, monthly 100 %; 50 requests a minute hard cap; no refunds |
| OpenCode Go: $10 a month over open coding models | A flat monthly allowance | Go $10, Go Plus $40, Go Teams $20 a member, 2-member minimum. Each model has its own limit: $1.50 on a $15 limit is 10 % of it | No per-token margin inside the allowance; past the limit, usage falls back to pay-as-you-go | 5-hour limit 20 % of the monthly allowance, weekly 50 %, resetting Monday 00:00 UTC; the client must send a session ID; Europe and the US by default |
| Alibaba Cloud Token Plan: a credit quota, not dollars | Monthly credit quotas on Qwen- and DeepSeek-class models | Lite $6, Essential $10, Standard $18, Pro $43 a month - $68 until 9 November 2026 - for 20,000 to 200,000 Credits | None published: a quota, not a rate table | New accounts only: $3 Lite and $5 Essential until 23:59:59 on 30 November 2026 (UTC+8); Singapore region only |
| Kimi Code membership: one quota across Kimi models | One membership shared by the CLI, VS Code, Desktop and other tools | Plus $19 a month ($15 annually), Pro $39, Max $99, Ultra $199; new members have no weekly quota | Extra Usage priced close to the Kimi Open Platform API rate, min top-up ¥25 | One shared quota, max 5 API keys; K3.1 Pro is thinking-only |
Where an aggregator breaks
Some margins do not depend on the model. OpenRouter's 5.5 % is charged when you buy credits and Openference's 5 % on every top-up; switching models touches neither. CometAPI's saving is published per model, so it exists only for the models on its list.
Pinning a provider changes the price, and some modes never sit at list price. NanoGPT publishes both exceptions: 5 % more when you pin a provider by hand, and bring-your-own-key at 5 % of the model cost. Its referral terms put caching mode, pinned providers, TEE and Private Mode, image and video through the API, paid search and research jobs outside eligible usage - none at plain text price.
Models and prices move. OpenCode Go and Openference say their model list may change as models are added. CometAPI's Black Friday rates cover 14 models and expire at 05:00 UTC on 3 December 2026; OpenCode Go's 50 % Gemini 4 Argon discount ends on 30 November 2026.
Reselling is where the terms get strict. Openference's terms say subscription allowances cover personal, interactive development and exclude commercial production, customer-facing API relays, background agent loops and headless daemons - enforced by inspecting request metadata. Alibaba Cloud's Token Plan is interactive use only: automation scripts, custom backends and batch calls are violations. Z.ai's GLM Coding Plan is personal-use only - a developer seat, not an API contract.
When an aggregator is worth it
- Count the tails, not the total. Ten models with small volumes is what one key solves; one model with a predictable bill wants a direct contract.
- Check that your model is on the price list. CometAPI's discount is published per model and its promotion names 14 models. A saving you cannot find on the list does not exist.
- Price the exceptions. Pinning a provider and bring-your-own-key cost extra at NanoGPT; past the allowance, Openference adds 5 % per top-up and OpenCode Go falls back to pay-as-you-go.
- Read the terms before the traffic arrives. Openference rejects unrecognised requests, Alibaba's key can be banned for batch calls, and a flat gateway is a budget that resets, not a cheaper token.
- Check the retention table if prompts are sensitive. OpenCode Go publishes retention per model: 0 days for GLM-5.3 and Kimi K3.1 Pro, 30 for Grok 4.7, 55 for Gemini 4 Argon.
Frequently asked questions
Why is an aggregator more expensive than the vendor list price? Because most aggregators buy credits in bulk, pass the provider rate through and charge for the purchase. OpenRouter's terms add 5.5 % to credit purchases with a $0.80 minimum, and Openference adds 5 % per top-up. Neither depends on the model.
Is it cheaper to pay an aggregator or the model vendor? It depends on the model and the volume. CometAPI's rate is 20 % below official on the models it prices. NanoGPT charges list price with no markup, so you pay the same and save the contract administration. Steady volume on one model favours a direct contract.
Can I resell access bought through an aggregator? Not on these terms. Openference's terms say subscription allowances are for personal, interactive development and exclude customer-facing API relays, background agent loops, batch pipelines and headless daemons. Alibaba Cloud's Token Plan bans automation scripts, custom backends and batch calls as violations.
What happens when a model leaves the catalogue? The catalogue is not a contract. OpenCode Go and Openference say the list may change as they test and add models, and CometAPI's promotion covers 14 models only. CometAPI says unused credits carry forward indefinitely; Openference says purchases are final.
Do aggregator APIs work with coding agents? Two of them do. OpenCode Go is built for coding agents and needs a stable session ID in the x-opencode-session header - apps without one cannot use Go models. Openference converts between OpenAI, Anthropic and Gemini formats and lists Cursor, Cline, Codex CLI and OpenCode as supported.
Sources
Vendor pages, read 19 November 2026 unless another date is given: CometAPI pricing and Black Friday · NanoGPT pricing and referrals (20 November) · OpenRouter FAQ · Openference pricing and terms (20 November) · OpenCode Go (19 November, re-read 20 November) · Alibaba Cloud notice and Token Plan overview · Kimi Code membership (20 November).
Referral links
- CometAPI - we get 20 % of every top-up, you get no discount - https://www.cometapi.com/?ref=evgeny&utm_source=media&utm_medium=article&utm_campaign=api-aggregators-aiincontext&utm_content=cometapi-referral
- NanoGPT - we get 10 % of eligible usage plus 5 % of subscriptions, you get 5 % off eligible usage - https://nano-gpt.com/r/yUf6yXwN?utm_source=media&utm_medium=article&utm_campaign=api-aggregators-aiincontext&utm_content=nanogpt-referral
- Z.ai - we get credits worth 10 % of a friend's first payment, you get 10 % off that first subscription - https://z.ai/subscribe?ic=FVQVPVL9L0&utm_source=media&utm_medium=article&utm_campaign=api-aggregators-aiincontext&utm_content=zai-referral
Affiliate disclosure: those three links are referral links; a qualifying CometAPI top-up may earn us a commission, and eligible NanoGPT usage or a first Z.ai subscription may earn us credits, at no cost to you. Every other link goes to the vendor and earns us nothing.
More from Elena Vasquez
Show full bio
Market editor covering technology and product news in artificial intelligence and machine learning at AI In Context.
795 articles
Related articles
- OpenAI ships GPT-6.2 Astra and Sol with 20% cheaper API pricing
- ChatGPT Dots require a $100-per-month Pro plan at launch
- OpenAI launches GPT-6.2 Astra and Sol, cuts API prices 20%
- ChatGPT Pro vs Claude Max vs Kimi: which AI subscription is worth it in 2026
- Subscription vs API: how to work out which one is cheaper when the model eats tokens