inference.net

免费大模型额度 · 免费模型 · 社区口碑 · 风险评估

网站类型官方站(免费层)
状态可用
口碑分76 / 100
风险等级
置信度72%
发现来源cheahjs/free-llm-api-resources

网站介绍

Inference.net(前身为分布式 GPU 网络 Kuzco)是主营付费的 AI 推理基础设施平台:提供 OpenAI 兼容的 serverless 开源模型 API(base URL api.inference.net/v1),并已把重心扩展到 Catalyst 网关/可观测/评测/训练全家桶;新用户有小额一次性赠金可用于按量调用模型。

免费额度

一次性新人赠金:上游目录 cheahjs/free-llm-api-resources 记录为「注册送 $1,回复邮件问卷再送 $25」;2026 年多篇第三方汇总也提到新账户约 $25 赠金可在 Playground/API 体验。赠金用完需充值,非按月重置。免费档速率限制 30 请求/分钟(付费档 250 RPM)。另有 Gateway 免费层(每月 100 万网关请求、1 席位、14 天数据保留),但那是路由/观测额度而非模型 token 额度。

限速/额度数据同步自 cheahjs/free-llm-api-resources(每日自动更新,非本站实测)

免费模型

meta-llama/llama-3.3-70b-instructmeta-llama/llama-3.1-8b-instructdeepseek/deepseek-v3-0324deepseek/deepseek-r1-0528openai/gpt-oss-120bopenai/gpt-oss-20bgoogle/gemma-3-27b-instructqwen/qwen3-30b-a3bqwen/qwq-32bmistralai/mistral-nemo-12b-instruct

优点

缺点

风险点

社区口碑综述

Inference.net 是由 Kuzco(Solana 上的分布式 LLM 推理 GPU 网络)演化而来的商业公司,采用节点信誉体系(验证/在线率/可靠性)调度任务。开发者社区讨论不算多,未见跑路、盗用 key 等负面记录;主要认知是「便宜的开源模型 serverless API + 新兴的 LLM 观测/训练平台」。作为送赠金的正规厂商风险低,但赠金小、且其分布式算力背景值得隐私敏感用户留意。

使用建议

适合想用一次性赠金低成本试用 Llama 3.3 70B / DeepSeek V3、R1 / gpt-oss 等开源大模型 OpenAI 兼容 API 的开发者;注册后记得完成邮件问卷拿 $25。不适合期待长期免费额度的用户;隐私敏感数据建议先阅读其数据保留文档。本条目仅作信息整理,不构成背书。

社区提及 (3)

cheahjs/free-llm-api-resources (GitHub) · neutral
Inference.net — Credits: $1, $25 on responding to email survey. Models: various open models.
docs.inference.net rate limits · neutral
Free tier: 30 requests per minute; paid tier: 250 requests per minute.
silv.blog — Distributed LLM Inference and the Rise of Kuzco · neutral
Kuzco is a distributed GPU cluster for LLM inference on Solana where users get paid to contribute idle compute, exposed via an OpenAI-compatible API (later rebranded/evolved into inference.net).

参考来源

In English

Summary: Inference.net (formerly the distributed GPU network Kuzco) is a paid-first AI inference infrastructure platform: it offers an OpenAI-compatible serverless API for open-source models (base URL api.inference.net/v1) and has expanded its focus to the Catalyst gateway/observability/evals/training suite; new users get a small one-time credit to spend on pay-per-token model calls.

Free quota: One-time signup credit: the upstream directory cheahjs/free-llm-api-resources records "$1 on signup, $25 on responding to an email survey"; several 2026 third-party roundups also mention roughly $25 in credits for new accounts to try the Playground/API. Credits are one-off, not monthly. Free tier is rate-limited to 30 requests/minute (paid tier 250 RPM). There is also a free Gateway plan (1M gateway requests/month, 1 seat, 14-day retention), but that is routing/observability quota, not model tokens.