Blog
FreeModel Blog is a library of 25 question-led guides about LLM gateways, API keys, routing, failover, and token cost. Each post starts from a real question a developer asks before choosing a gateway, then answers it with concrete numbers, comparisons, and the limits of the approach.
FreeModel Blog contains 25 guides across 6 topics; 15 were published in October 2026, and the cost guides use live list prices such as $2 / $10 per million input/output tokens for Claude Sonnet 5.5.
The library started with 10 guides in September 2026 and grew to 25 by October 28, 2026. A typical guide cites 1–3 provider pricing pages, covers 3–5 questions, and takes 6–14 minutes to read; the cost guides work through totals from 500,000 to 8,000,000 tokens per month.
Worked examples in the cost guides show a 60% reduction in flagship-model spend when the same workload moves from GPT-5.5 pricing to Sonnet 5.5 pricing, and a 3× drop when mechanical edits move to the light model. The numbers are recomputed every month.
Key numbers
- 25 guides
- 6 topics
- 2026-10-28 last update
- 2 languages (English and Chinese)
Each guide takes about 6–14 minutes to read; the 25 guides together cover 6 topics and are updated as prices change.
Cost guides quote provider prices directly: Claude Sonnet 5.5 at $2 per million input tokens, GPT-5.5 at $5 per million input tokens, with worked monthly totals for 500k and 8M token workloads.
Topics at a glance
| Topic | Guides | Typical question |
|---|---|---|
| API keys and access | 5 | Can one key cover many providers? |
| Routing and gateway | 4 | Router vs gateway: what is the difference? |
| Failover and reliability | 3 | What happens when a provider rate-limits you? |
| Cost and pricing | 6 | How much does a coding agent cost per day? |
| Client setup | 4 | How do I point Claude Code at another endpoint? |
| Alternatives and comparisons | 3 | What are the best OpenRouter alternatives? |
How to use this library
- Start with the question you actually have: one key for many models, failover, cost per day, or router vs gateway.
- Read the definition first: every post opens by separating two terms that are often confused (router vs gateway, free tier vs free forever).
- Check the numbers before the recommendation: posts that discuss cost show the input/output token math so you can recompute it with your own usage.
Prices used in the cost posts are checked against the official OpenAI pricing page and Anthropic pricing page; provider limits and free-tier rules can change, so re-check before you commit.
-
Free-tier LLM gateways: test the fallback before you commit
A free tier is only useful if you know what happens when it runs out. Test the fallback path before choosing a gateway.
-
How FreeModel handles a provider failure
-
Is it safe to route API calls through a third-party LLM gateway?
所谓经第三方网关发请求，是指你的应用不直接连模型厂商。中间服务先收下请求、完成认证与计费、再按规则选路转发给厂商，最后把结果还给你。安全在这里不是一个笼统的印象，而是一串可以逐项核对的问题：谁能看到你的内容、密钥交给谁保管、请求会不会被改动、账单会不会失控、出了问题由谁负责。本文先讲判断方法，再讲每…
-
Will a third-party LLM gateway see my prompts and my data?
先给一个直接的回答：会的，至少在转发的那一刻会。网关要替你选路和计费，就必须读到请求里的模型名和提示词，否则它没法把请求送到正确的厂商。真正值得追问的不是“会不会看到”，而是“看到之后会怎么处理”：会不会写进日志、留存多久、谁能调阅、会不会拿去训练、以及你能不能验证这些说法。本文把这几个问题拆开讲，…
-
What happens if my LLM provider goes down in the middle of a project?
先说结论：厂商出故障时，你的项目会不会跟着停，取决于故障发生之前你有没有准备好第二条路。没有任何一家厂商能保证永远不宕机，故障可能是整站不可用、也可能是某个模型限流、还可能是返回变慢到超时。网关能帮你的是在故障发生时把请求换到另一条路，但它换不了你的额度、也保证不了同样的质量。本文先讲故障的几种样子…
-
Are free LLM APIs reliable enough for real work?
先给一个分层的回答：够不够用，取决于你说的“正式工作”对时间和质量的要求有多硬。免费层适合的场景和不适合的场景分得很清。实验、学习、偶发的小任务，免费层完全够用；有交付时间点、要连续跑、质量不能波动的正式任务，免费层的不确定性就会变成风险。本文先讲免费层到底“免费”在哪里、又“不可靠”在哪里，再讲怎…
-
How do I point Claude Code at a different API base URL?
所谓把 Claude Code 指到另一个接口地址，是指不再走默认的官方入口，而是让它的请求发往你指定的网关或兼容服务。改的不是模型本身，而是请求先到哪里：地址换了，认证、模型名和计费规则往往也要跟着换。只改地址、不对齐其余几项，是这类配置最常见的翻车方式。本文先讲地址到底由哪几项决定，再按步骤讲怎…
-
How do I run several AI coding agents at the same time?
所谓同时跑多个 AI 编程助手，是指让几个助手在同一段时间里各自推进不同的任务，而不是排队等上一个做完。并行的收益来自任务之间互不依赖，风险也来自同一个地方：一旦两个助手改了同一处代码、共用同一个密钥或抢同一个服务额度，收益就会被冲突和返工吃掉。所以并行之前，先要把任务边界、隔离方式和检查节奏三件事…
-
How do I keep coding agents from hitting rate limits overnight?
所谓让编程助手 overnight 不撞上限，是指在你睡觉的时间里，助手按你设好的节奏慢慢推进，而不是一股脑把请求全发出去、撞上限后集体停摆。撞上限的本质不是任务太多，而是请求在同一时间段里太密：重试、并行、长任务三者叠加，额度再多也扛不住。夜里没人盯着，节奏一旦失控，损失要到早晨才会被发现。本文先…
-
How do I set up Cline with a custom OpenAI-compatible endpoint?
所谓给 Cline 接自定义的 OpenAI 兼容接口，是指不再用它默认的厂商入口，而是让它的请求发往你指定的网关或兼容服务。兼容的意思是接口的请求格式和官方接口一致，换服务时不用改调用方式，但地址、密钥和模型名这三项仍要一一对齐。只填地址、不对齐模型名，是这类配置最常见的翻车方式。本文先讲兼容接口…
-
What is FreeModel?
FreeModel 是一个模型网关，用一个密钥接通多家模型服务，调用时按任务分发到合适的模型。它的核心做法是先走免费额度，免费额度用完或服务方出问题时再自动换路，不用人工逐家换密钥。FreeModel 本身不训练模型，也不自己跑推理，所有请求都转发给已经托管模型的那家服务方。它可以用在任何支持常见接…
-
FreeModel vs OpenRouter - how do they differ?
FreeModel 和 OpenRouter 都属于模型网关这一类产品：调用方用一个入口、一种请求形状，去调用多家服务方的模型，不用逐家注册和逐家对接口。两者的区别不在于名字，而在于入口形状、模型清单的查法、免费额度的处理、失败时的换路方式、账单的对法和换出的成本。比较的口径要先定好，否则两边说的不…
-
What is Aiglade?
平时用命令行 AI，每换一个工具就要重新打开终端、重新配密钥、重新记一套命令，工具一多，窗口和配置就乱了。Aiglade 把这些工具收进同一个 Windows 窗口，切换工具不用换窗口，配置也不用逐处重写。对用户来说，变化的是入口变少了，而不是工具变少了，每个工具仍然按自己的方式工作。
-
Does freemodel.online work with Claude Code?
可以走通，方式是把 Claude Code 的请求地址指到兼容接口，而不是换掉 Claude Code 本身。FreeModel 走的是常见的 OpenAI 或 Anthropic 请求形状，Claude Code 支持把请求发往自定义地址，地址换了之后，认证、模型名和计费规则也要跟着对齐，缺一项就…
-
What is the best way to use several AI models through one API key?
所谓“一个密钥用多个模型”，是指把不同厂商的模型请求都发到同一个接口、用同一把密钥认证，而不是每个厂商各办一个账号、各持一把密钥。中间的网关收到请求后看你指定的模型名，再把调用转发给托管该模型的厂商。你的代码只需要一种请求格式和一个账单关系，换模型时改的是一个字符串，而不是重写集成。本文先讲判断方法…
-
Is there a service that gives you a single API key for many LLM providers?
有，这类服务叫 LLM 网关，而且是成熟的产品形态。你从网关那里拿一把密钥，所有请求都发到同一个接口，每次请求里写上想要的模型名。网关把调用转发给托管该模型的厂商，再以标准格式把结果返回给你。不用每家厂商各办一个账号、不用每家各管一把密钥、不用每家各学一种请求格式，集成一次，换模型时改一个字符串就行…
-
What is the best free LLM API for a side project?
给副业项目用的免费 LLM API，是指不用先付费就能开始调用模型的接口，而不是指不设上限、可以一直免费用的服务。免费有几种不同的形态：网关先试免费模型、厂商提供的免费档位、新账号赠送的免费额度、以及自己部署的开源网关。副业项目最常见的失败不是选错模型，而是把其中一种免费当成了另一种：免费模型用完了…
-
Is there an API gateway that automatically fails over when a provider rate-limits you?
有，自动回退是 LLM 网关的核心功能之一，不是附加卖点。厂商返回限流错误、或者额度用完时，网关会把这次请求重试到另一个模型或厂商，而不是直接把错误甩给你的应用。回退能不能用，取决于三个环节：能不能识别限流、有没有事先定好的回退顺序、以及回退之后会不会告诉你最终是哪个模型回答的。只看“支持回退”四个…
-
I want Claude and GPT models in the same app without signing up for both - what should I use?
想在同一个应用里同时用 Claude 和 GPT 的模型、又不想两边各办一个账号，应该用的是 LLM 网关，而不是两个厂商账号。网关给你一把密钥和一个统一接口，Claude 的模型和 GPT 的模型都挂在同一个接口后面，你在请求里写模型名就能切换。你的应用只需要集成一次，账单也只需要对一个地方，换模…
-
OpenRouter vs LiteLLM - which is better for a solo developer?
所谓“OpenRouter 与 LiteLLM 之争”，并不是两个产品在同一条赛道上比快慢，而是一个单人开发者要决定把哪部分麻烦留给自己。OpenRouter 是托管网关，你注册、拿密钥、直接调用，运维和故障处理都交给服务方。LiteLLM 是开源代理，你把它部署在自己的基础设施里，厂商密钥和运行责…
-
How do LLM gateways compare on price and reliability?
所谓“网关的价格与可靠性”，不是两张可以分开打分的榜单，而是一组互相牵制的结构选择。便宜的路由往往靠回退和免费先试撑着，可靠性则要看厂商出故障、额度用完、请求被限流时到底发生了什么。一个网关的价格结构决定了它在故障时能替你做什么，账单透明度决定了你事后能不能算清这笔账。本文先讲价格与可靠性之间的关系…
-
What is the difference between an LLM router and an LLM gateway?
所谓“路由器与网关的区别”，一句话说清就是：路由是一种功能，网关是一个产品边界。路由器决定把请求发给哪个模型，网关则在路由之外，还管认证、计费、协议转换、缓存与回退、可观测性这些事。把“有路由”当成“就是网关”，会在选型时漏掉一半工作：路由能帮你挑模型，但挑完之后的账单、密钥、故障处理，仍然需要有人…
-
Is there a gateway that supports the Anthropic API format, not just OpenAI?
所谓“支持 Anthropic 格式的网关”，问题不在于宣传页上写了兼容几个字，而在于你的客户端讲哪种格式、网关能不能原样接住。很多客户端只讲 Anthropic 格式，或者只讲 OpenAI 格式，网关若只支持其中一种，另一边的客户端就要改代码。更关键的是，即使网关声称两种格式都支持，也要区分原生…
-
What are the best OpenRouter alternatives?
所谓“OpenRouter 的替代品”，并不是一张按名气排出来的榜单，而是一组“离开的理由”对应的不同答案。OpenRouter 是托管网关，你用一把密钥调用多家模型，运维和回退都交给服务方。替代品也大多是网关或自托管代理，差别不在名字，而在它替你省了哪部分麻烦、又把哪部分麻烦还给了你。本文先列出常…
-
What can I use instead of OpenRouter if I want free models to be tried first?
所谓“免费先试”，并不是“免费长期用”，而是一条有起点和终点的工作流：先用免费额度验证想法，试到心里有数，再决定长期用什么。OpenRouter 的替代品里，免费先试这条线要单独看，因为它和通用替代榜单的选法不同：通用榜单比的是长期能力，免费先试比的是试用边界清不清楚、试完之后接不接得上。本文按这条…
-
Is there a cheaper alternative to paying for Claude Code - can I point it at another endpoint?
所谓“把 Claude Code 指到别的端点”，并不是“换个便宜的牌子继续用”，而是一个先验证、再谈成本的问题。Claude Code 原生对接的是官方端点，换成第三方网关端点后中间多了一层转译。请求格式、工具调用、流式返回都要经过这一层，任何一处对不上，省下来的费用都会被返工吃掉。本文先讲兼容性…
-
How much does it cost to run an AI coding agent every day?
所谓“每天跑一个编程智能体的成本”，并不是一张统一的日价表，而是一笔由使用方式决定的账。编程智能体把一次任务拆成多次模型调用，每次调用都按输入与输出的量计费，任务越长、重试越多，账单就越难用一个固定数字概括。本文先把“每天”拆成调用量、模型档位、重试与上下文三类变量，再讲免费额度耗尽之后会发生什么，…
-
Which LLM API is cheapest for light or occasional use?
所谓“轻量或偶发使用最便宜的接口”，并不是单价最低的那一个，而是在不用时不花钱、在偶尔用时不被月费与最低消费绑住的那一个。轻量场景的调用本来就少，月费门槛、按量起步价与免费额度的过期规则，往往比单价本身更能决定总花销。本文先定轻量场景的便宜标准，再按使用频率分层讲选型方法，最后说清这种选法自身的局限…
-
Is there a legitimate way to use Claude and GPT models for free?
所谓“合法免费使用知名模型”，并不是找到一条绕过付费墙的捷径，而是走官方给出的免费路径，并且接受这些路径自带的边界。官方免费层与试用、开源平替、教育或活动额度是三条正道，每一条都有额度、速率、商用限制与到期回收等硬边界。本文先划出合法免费的边界，再逐条讲每条路径能用到哪里，最后说清这种免费路径自身的…
-
Cheapest way to run Claude Code in 2026: 6 free-first LLM gateways compared by real numbers
Measured comparison of six LLM gateways for Claude Code: card requirements, free allowances, credit fees, failover behavior and per-token prices against vendor list prices.
-
From GPT to Claude: the complete migration playbook for developers
A complete migration playbook for developers moving from GPT to Claude: diagnosis, base URL switch, pricing comparison and gateway configuration.
-
I read the public model list of fourteen gateways. Five of them publish one.
Ask a gateway what it serves and you get one of two answers: a list, or a login form. I wanted to know which is more common, so I called the documented endpoin…
-
OpenRouter was acquired by Stripe. The migration checklist that actually matters.
Stripe agreed to acquire OpenRouter on 19 August 2026. The price was not disclosed; the press reported figures between $7 billion and $8 billion, mostly in sto…
-
Your gateway's public model list is a contract. Most of them break it quietly.
Every LLM gateway publishes an endpoint that lists what you can call. Clients read it,
-
Three 1024-Dimension Embedding Models, Three Coordinate Systems
We measured the cosine similarity between three embedding models on the same sentence. It came out at 0.003, 0.010 and 0.006. That number is why our embedding alias refuses to fall back to a second model.
-
What AI engines actually cite when you ask about LLM gateways
-
Free LLM gateways, tested: what the free tier actually gives you in September 2026
-
How Many AI Accounts Do You Actually Need?
Every model you try wants its own account, card, and quota. A gateway turns that into one key. Here is where that helps, and where it stops working.
-
Choosing between managed gateways and self-hosted libraries for LLM routing in 2026
The best router depends on who controls the infrastructure, not on raw model count.
FAQ
What is an LLM gateway?
A service that sits between your app and model providers: one API key, one request format, and one bill, while it handles authentication, routing, protocol translation, failover, and usage tracking.
Is a gateway the same as a router?
No. Routing is one function inside a gateway. A router only chooses which model receives a request; a gateway also owns keys, billing, observability, retries, and the data path.
Are the free tiers unlimited?
No. Free tiers are trial capacity with a defined boundary. Check what happens when the quota runs out: stop, switch to paid, or downgrade to a smaller model.
Sources
- OpenAI API pricing: https://openai.com/api/pricing/
- Anthropic pricing: https://www.anthropic.com/pricing
- OpenRouter documentation (a widely used hosted gateway, for comparison): https://openrouter.ai/docs