炸裂!商汤Token Plan里提供免费GLM-5.2Explosive! SenseTime's Token Plan offers free GLM-5.2


AI free token/AI免费积分

本文来源:https://mp.weixin.qq.com/s/4f4zekqfQmTduPRBx9XSYQ
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

最近主力模型账单蹭蹭涨,我就想找个免费模型打辅助——日常文档、报告总结这些活儿,没必要全烧付费token。

翻到商汤的日日新SenseNova,Token Plan公测,0元/月。我第一反应:啊?真的假的。注册走了一遍,调用跑通了,钱确实不收。

4f2b4171d13df1238290cda679540ab9.png

每个模型每5小时能调1500次,一个账号最多建20个API Key。没预充值、没到期作废,注册就能用。个人开发者做原型,这个量够做出个demo了。

模型清单,有个彩蛋

公测放开的主要是这几款:

  • DeepSeek V4 Flash:深度求索的旗舰推理模型,逻辑、算账、写代码、长文分析都行。
  • SenseNova 6.7 Flash-Lite:商汤自研原生多模态,读文档、图片、表格,跑办公长链路。
  • SenseNova U1 Fast:专门出PPT和信息图,排版能直接交初稿。

重点是第四个——我在模型列表里拉到了GLM-5.2。

微信图片_2026-07-08_122930_965.png

我实测了两把

第一把,拿GLM-5.2诊断我的小程序项目,几十个文件,等了大概10分钟出报告。是有点慢,但一想免费,我忍了。报告不是水货——它真读完了项目结构,按严重程度分类,还给了修复优先级。比之前几款国产模型跑同一个项目,详尽度确实更胜一筹。

第二把,拿6.7 Flash-Lite把10份月度Excel汇总成绩效报告,从字段清洗到图表说明一条链路跑完,中间没让我手动拼装。

怎么接,几行代码

进 platform.sensenova.cn 注册,控制台建API Key。关键就三个:

  • Base URL:https://token.sensenova.cn/v1
  • 模型名:比如 sensenova-6.7-flash-lite(GLM-5.2用列表里的实际名)
  • API Key:控制台那个

SenseNova兼容OpenAI协议,Python直接调:

from openai import OpenAIclient = OpenAI(api_key="你的KEY", base_url="https://token.sensenova.cn/v1")r = client.chat.completions.create(model="sensenova-6.7-flash-lite", messages=[{"role":"user","content":"你好"}])print(r.choices[0].message.content)

Cursor、Claude Code这类吃OpenAI协议的工具,填上面信息就能接。

日常办公、批量任务、轻量编程、内容生产,报告汇总、数据洞察、PPT初稿,这套免费额度很大方。但高精度推理(法律、金融风控)跟顶级旗舰比还有差距,U1 Fast的PPT复杂排版偶尔翻车,得人工过一道。

GLM-5.2这边的额度官网没明说,推测不会太高,类比DeepSeek大概不超500次/5h。付费档Lite和Pro已经预告"即将上线",公测免费不会一直持续。趁免费,先跑一波再说。

 

Source: https://mp.weixin.qq.com/s/4f4zekqfQmTduPRBx9XSYQ If there are copyright issues, please let us know via email: 13951431913#139.com or submit a comment on the right.

My main model bill has been rising recently, so I was looking for a free model to help — for daily document reading and report summaries, I don’t need to burn paid tokens.

I came across SenseNova’s Token Plan, zero yuan per month during public beta. My first reaction: is this for real? I went through the registration and ran a test call — it worked, and they didn’t charge me.

Each model allows up to 1,500 calls every 5 hours, and one account can create up to 20 API keys. No prepayment, no expiration — you can start using it right after registering. For a solo developer building a prototype, this volume is enough to produce a demo.

The model list held a surprise

The main models released for public beta are:

DeepSeek V4 Flash: DeepSeek’s flagship inference model, good for logic, calculations, coding, and long-form analysis.

SenseNova 6.7 Flash-Lite: SenseTime’s native multimodal model, reads documents, images, tables — supports long office workflows.

SenseNova U1 Fast: Specialized for PPT and infographs, can produce draft-ready layouts.

The key fourth entry — I scrolled down and found GLM-5.2 in the list.

I ran two tests

I ran the first test with GLM-5.2 on my miniprogram project, containing dozens of files. The report took about 10 minutes to generate — a bit slow, but since it’s free, I’m not complaining. The analysis was solid: it accurately mapped the project structure, classified issues by severity, and provided clear repair priorities. Compared to previous domestic models I’d tried on the same project, the depth and accuracy were notably better.

Second, I used 6.7 Flash-Lite to process a batch of 10 monthly Excel reports — from field cleaning to chart annotation, all in one go. No need for manual assembly steps.

How to get started, in a few lines of code

Register at platform.sensenova.cn, create an API key in the console. You’ll need three things:

Base URL: https://token.sensenova.cn/v1

Model name: e.g., sensenova-6.7-flash-lite (for GLM-5.2, use the actual name from the list)

API Key: from the console

SenseNova supports OpenAI protocol, so you can call it directly from Python:

import requests headers = {"Content-Type": "application/json", "Authorization": "Bearer YOUR_API_KEY"} data = {"model": "sensenova-6.7-flash-lite", "messages": [{"role": "user", "content": "What is qubit coherence time?"}]} response = requests.post("https://token.sensenova.cn/v1/chat/completions", headers=headers, json=data) print(response.json()["choices"][0]["message"]["content"])

Tools like Cursor and Claude Code that support OpenAI protocol can be configured with this information — no need to rewrite plugins.

This free quota is generous for daily office work, batch tasks, light programming, and content production — think report summaries, data insights, and PPT drafts. However, for high-precision inference (like legal or financial risk analysis), it doesn’t yet match top-tier flagships. Also, U1 Fast’s PPT layouts can sometimes be off, so manual review is still needed.

4f2b4171d13df1238290cda679540ab9.png

微信图片_2026-07-08_122930_965.png

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry