中文读者:直达中文部分 · 中文 Top 15 · 中文评分说明 | English: Top 15 · How the score works
Tokenation collects free LLM API offers and checks every entry against the provider's own pages. Offers are split into two kinds:
Evergreen free tiers: standing free quotas with no announced end date, plus new-account trial allowances.
Each row records the provider, the model ID, a one-line overview, the free quota, the deadline and the score.
Limited-time offers: announced by the provider as free for a limited time, either with a fixed end date
or with no published deadline (the entry says so). Offers whose announced deadline has already passed are removed.
The 10 best deals on the site right now, ranked by the Tokenation score.
| Provider | Type | Model ID | Overview | Free quota | AA | Free until | Score | Link |
|---|---|---|---|---|---|---|---|---|
| Google Gemini API | Evergreen | gemini-4-hydrogen |
Google's Gemini 4 flagship with a 1M-token context | Gemini 4 Hydrogen and 3.8 Flash… | 50.4 | Ongoing | 93.61 | → |
| OpenCode Zen | Limited-time | deepseek-v4.1-flash-free |
DeepSeek V4.1 Flash free for a limited time | 12 limited-time free models; OpenCode… | 32.6 | No deadline announced | 81.73 | → |
| OpenCode Go | Limited-time | beam-free |
Beam open-weight model, 501B total | 4 limited-time free models; outside the… | 30.3 | No deadline announced | 80.20 | → |
| Vercel AI Gateway | Limited-time | reflection/beam-free |
Beam open-weight model with ~501B total parameters | 4 readable models at $0 in/out (28… | 30.3 | No deadline announced | 80.20 | → |
| Google Gemini API | Evergreen | gemini-3.8-flash |
Google's strongest Flash model for coding, agents and multi-step reasoning. | Gemini 4 Hydrogen and 3.8 Flash… | 30.8 | Ongoing | 68.55 | → |
| AutoClaw (Z.ai / Zhipu) | Limited-time | glm-5.3 |
Z.ai reasoning model for software engineering and long-horizon agent tasks. | 50M intl / 100M CN tokens; Nov 29 | 31.7 | 2026-11-29 (over 7 days left) | 57.15 | → |
| OpenRouter | Evergreen | thinkingmachines/inkling-2-small:free |
Multimodal MoE reasoning model from Thinking Machines. | :free models $0, 20 RPM, 50 RPD | 30.6 | Ongoing | 54.57 | → |
| AiHubMix | Evergreen | xiaomi-mimo-v2.6-pro-free |
Xiaomi MiMo V2.6 Pro flagship | 76 models at $0; 100 req/day, 1M tokens… | 46.3 | Ongoing | 53.47 | → |
| AiHubMix | Evergreen | qwen3.8-max-free |
Qwen3.8 Max flagship | 76 models at $0; 100 req/day, 1M tokens… | 45.4 | Ongoing | 52.87 | → |
| AiHubMix | Evergreen | coding-glm-5.3-free |
GLM 5.3 coding variant | 76 models at $0; 100 req/day, 1M tokens… | 44.8 | Ongoing | 52.47 | → |
Top 10 of 83 scored model rows. Scoring rule below. Full lists below.
Every offer here carries one number from 0 to 100, so an offer you have never touched can be compared against one you already pay for. Free API tiers change every week, and signing up to test each one costs a weekend, so the score compresses that decision into a single number. It measures how much usable inference you get for free right now. It is not a ranking of model quality.
-
Intelligence, 40%: the model's Artificial Analysis index.
-
Free quota, 60%: how much you can actually run before you hit a wall.
-
Every score traces back to the provider's own page, rechecked daily.
-
Out of scope, so never scored: decision models that return typed answers instead of text, routing endpoints (an "auto" model selection), and models under 20B total parameters.
↑ Back to top · Chinese version
Tokenation 每日搜集免费 LLM API,并逐条回到厂商官方页面核实。条目分两类:
常态免费(Evergreen),官方定价页长期挂着、无公告结束时间的免费额度,含新用户免费额度。
每行记录平台、模型 ID、一段简介、免费额度、截止日与推荐指数。
限时免费(Limit-time),官方明示的限时免费 offer,含两类:写明固定截止日的,以及只写「限时」
未公布截止日的(会在日期字段注明)。已过公布截止日的条目一律移除。
本站此刻最值的 10 单,按推荐指数从高到低。
| 平台 | 类型 | 模型 ID | 介绍 | 免费额度 | AA | 截止 | 推荐指数 | 链接 |
|---|---|---|---|---|---|---|---|---|
| Google Gemini API | 常态 | gemini-4-hydrogen |
谷歌 Gemini 4 旗舰,统一 1M 上下文 | Gemini 4 Hydrogen、3.8 Flash… | 50.4 | 长期有效 | 93.61 | → |
| OpenCode Zen | 限时 | deepseek-v4.1-flash-free |
DeepSeek V4.1 Flash 限时免费模型 | 12 个限时免费模型;需在 OpenCode 内使用 | 32.6 | 未公布截止日 | 81.73 | → |
| OpenCode Go | 限时 | beam-free |
Beam 开放权重模型,501B 总参数 | 4 个限时免费模型;不计入 Go 月度额度 | 30.3 | 未公布截止日 | 80.20 | → |
| Vercel AI Gateway | 限时 | reflection/beam-free |
Beam 开放权重模型,约 501B 总参数 | 4 个可读模型输入输出 $0(共 28 个促销位) | 30.3 | 未公布截止日 | 80.20 | → |
| Google Gemini API | 常态 | gemini-3.8-flash |
谷歌最强 Flash,擅长编码、Agent 与多步推理 | Gemini 4 Hydrogen、3.8 Flash… | 30.8 | 长期有效 | 68.55 | → |
| AutoClaw(智谱) | 限时 | glm-5.3 |
智谱推理模型,面向软件工程与长周期 Agent 任务 | 国际站 5,000 万、中文站 1 亿 Tokens;11/29 到期 | 31.7 | 2026-11-29(剩余超7天) | 57.15 | → |
| OpenRouter | 常态 | thinkingmachines/inkling-2-small:free |
Thinking Machines 多模态 MoE 推理模型 | :free 模型 $0,20 RPM、50 RPD | 30.6 | 长期有效 | 54.57 | → |
| AiHubMix | 常态 | xiaomi-mimo-v2.6-pro-free |
小米 MiMo V2.6 Pro 旗舰 | 76 款模型 $0;充值后每日 100 请求、1M tokens | 46.3 | 长期有效 | 53.47 | → |
| AiHubMix | 常态 | qwen3.8-max-free |
通义 Qwen3.8 Max 旗舰 | 76 款模型 $0;充值后每日 100 请求、1M tokens | 45.4 | 长期有效 | 52.87 | → |
| AiHubMix | 常态 | coding-glm-5.3-free |
GLM 5.3 编码版 | 76 款模型 $0;充值后每日 100 请求、1M tokens | 44.8 | 长期有效 | 52.47 | → |
上表从 83 行已评分条目中取前 10 行,评分口径见下节。完整清单见下方四个列表。
本站每个条目都带一个 0 到 100 的分数,让你能拿一个没试过的免费额度,和一个已经在用的放在一起比。免费额度每周都在变,逐个注册去试要花掉一个周末,这个分数把这件事压缩成一个数字。它量的是此刻免费能拿到多少可用推理,不是模型强弱排名。
-
智能占 40%:取模型在 Artificial Analysis 智能指数上的位置。
-
免费额度占 60%:真跑起来撞墙之前能用多少。
-
每个分数都能追回厂商自己的页面,每天重新核实一次。
-
不参与评分:只返回类型化结果的决策模型、路由端点(auto 自动选模),以及总参数低于 20B 的模型。